Intelligent Human Computer Interaction

10th International Conference IHCI 2018

Allahabad India December 7 9 2018
Proceedings Uma Shanker Tiwary
Visit to download the full and correct content document:
https://textbookfull.com/product/intelligent-human-computer-interaction-10th-internatio
nal-conference-ihci-2018-allahabad-india-december-7-9-2018-proceedings-uma-shan
ker-tiwary/
More products digital (pdf, epub, mobi) instant
download maybe you interests ...

Intelligent Human Computer Interaction 8th

International Conference IHCI 2016 Pilani India
December 12 13 2016 Proceedings 1st Edition Anupam Basu

https://textbookfull.com/product/intelligent-human-computer-
interaction-8th-international-conference-ihci-2016-pilani-india-
december-12-13-2016-proceedings-1st-edition-anupam-basu/

Intelligent Human Computer Interaction 9th

International Conference IHCI 2017 Evry France December
11 13 2017 Proceedings 1st Edition Patrick Horain

https://textbookfull.com/product/intelligent-human-computer-
interaction-9th-international-conference-ihci-2017-evry-france-
december-11-13-2017-proceedings-1st-edition-patrick-horain/

Brain Informatics International Conference BI 2018

Arlington TX USA December 7 9 2018 Proceedings Shouyi
Wang

https://textbookfull.com/product/brain-informatics-international-
conference-bi-2018-arlington-tx-usa-
december-7-9-2018-proceedings-shouyi-wang/

Intelligent Human Systems Integration Proceedings of

the 1st International Conference on Intelligent Human
Systems Integration IHSI 2018 Integrating People and
Intelligent Systems January 7 9 2018 Dubai United Arab
Emirates 1st Edition Waldemar Karwowski
https://textbookfull.com/product/intelligent-human-systems-
integration-proceedings-of-the-1st-international-conference-on-
intelligent-human-systems-integration-ihsi-2018-integrating-
Advances in Data Science Third International Conference
on Intelligent Information Technologies ICIIT 2018
Chennai India December 11 14 2018 Proceedings Leman
Akoglu
https://textbookfull.com/product/advances-in-data-science-third-
international-conference-on-intelligent-information-technologies-
iciit-2018-chennai-india-december-11-14-2018-proceedings-leman-
akoglu/

Big Data Analytics 6th International Conference BDA

2018 Warangal India December 18 21 2018 Proceedings
Anirban Mondal

https://textbookfull.com/product/big-data-analytics-6th-
international-conference-bda-2018-warangal-india-
december-18-21-2018-proceedings-anirban-mondal/

Information Systems Security: 14th International

Conference, ICISS 2018, Bangalore, India, December
17-19, 2018, Proceedings Vinod Ganapathy

https://textbookfull.com/product/information-systems-
security-14th-international-conference-iciss-2018-bangalore-
india-december-17-19-2018-proceedings-vinod-ganapathy/

Communication Systems and Networks: 10th International

Conference, COMSNETS 2018, Bangalore, India, January
3-7, 2018, Extended Selected Papers Subir Biswas

https://textbookfull.com/product/communication-systems-and-
networks-10th-international-conference-comsnets-2018-bangalore-
india-january-3-7-2018-extended-selected-papers-subir-biswas/

Similarity Search and Applications 11th International

Conference SISAP 2018 Lima Peru October 7 9 2018
Proceedings Stéphane Marchand-Maillet

https://textbookfull.com/product/similarity-search-and-
applications-11th-international-conference-sisap-2018-lima-peru-
october-7-9-2018-proceedings-stephane-marchand-maillet/
 Uma Shanker Tiwary (Ed.)

Intelligent
LNCS 11278

Human Computer
Interaction
10th International Conference, IHCI 2018
Allahabad, India, December 7–9, 2018
Proceedings

123
Lecture Notes in Computer Science 11278
Commenced Publication in 1973
Founding and Former Series Editors:
Gerhard Goos, Juris Hartmanis, and Jan van Leeuwen

Editorial Board
David Hutchison
Lancaster University, Lancaster, UK
Takeo Kanade
Carnegie Mellon University, Pittsburgh, PA, USA
Josef Kittler
University of Surrey, Guildford, UK
Jon M. Kleinberg
Cornell University, Ithaca, NY, USA
Friedemann Mattern
ETH Zurich, Zurich, Switzerland
John C. Mitchell
Stanford University, Stanford, CA, USA
Moni Naor
Weizmann Institute of Science, Rehovot, Israel
C. Pandu Rangan
Indian Institute of Technology Madras, Chennai, India
Bernhard Steffen
TU Dortmund University, Dortmund, Germany
Demetri Terzopoulos
University of California, Los Angeles, CA, USA
Doug Tygar
University of California, Berkeley, CA, USA
Gerhard Weikum
Max Planck Institute for Informatics, Saarbrücken, Germany
More information about this series at http://www.springer.com/series/7409
Uma Shanker Tiwary (Ed.)

Intelligent
Human Computer
Interaction
10th International Conference, IHCI 2018
Allahabad, India, December 7–9, 2018
Proceedings

123
Editor
Uma Shanker Tiwary
Indian Institute of Information Technology
Allahabad, India

ISSN 0302-9743 ISSN 1611-3349 (electronic)

Lecture Notes in Computer Science
ISBN 978-3-030-04020-8 ISBN 978-3-030-04021-5 (eBook)
https://doi.org/10.1007/978-3-030-04021-5

Library of Congress Control Number: 2018960863

LNCS Sublibrary: SL3 – Information Systems and Applications, incl. Internet/Web, and HCI

© Springer Nature Switzerland AG 2018

This work is subject to copyright. All rights are reserved by the Publisher, whether the whole or part of the
material is concerned, specifically the rights of translation, reprinting, reuse of illustrations, recitation,
broadcasting, reproduction on microfilms or in any other physical way, and transmission or information
storage and retrieval, electronic adaptation, computer software, or by similar or dissimilar methodology now
known or hereafter developed.
The use of general descriptive names, registered names, trademarks, service marks, etc. in this publication
does not imply, even in the absence of a specific statement, that such names are exempt from the relevant
protective laws and regulations and therefore free for general use.
The publisher, the authors, and the editors are safe to assume that the advice and information in this book are
believed to be true and accurate at the date of publication. Neither the publisher nor the authors or the editors
give a warranty, express or implied, with respect to the material contained herein or for any errors or
omissions that may have been made. The publisher remains neutral with regard to jurisdictional claims in
published maps and institutional affiliations.

This Springer imprint is published by the registered company Springer Nature Switzerland AG
The registered company address is: Gewerbestrasse 11, 6330 Cham, Switzerland
 Preface

It is indeed a great privilege to present the proceedings of the 10th International

Conference on Intelligent Human Computer Interaction (IHCI 2018) organized by the
Indian Institute of Information Technology, Allahabad during December 07–09, 2018.
This is one of the rare conferences focusing on human–computer interaction (HCI).
The conference aims to explore new dimensions of HCI science and technologies
instead of confining HCI only to various types of interfaces.
In total, 89 papers were submitted. The review committee had a very challenging
task of selecting high-quality submissions. Each paper was peer-reviewed by at least
three reviewers. On the recommendation of the Program Committee, 28 papers were
accepted for presentation at the main conference.
The proceedings are organized in five chapters corresponding to each track of the
conference as described below:
The ‘ECG/EEG-Based and Other Multimodal Interactions’ track was related to
enabling technology that encompasses brain–computer interfaces, body sensors and
communication multimodal interfaces, ECG/EEG-based feedback and control for HCI,
ECG/EEG-based emotion, affect analysis and detection. In this track, eight papers were
accepted. The ‘Natural Language, Speech and Dialogue Processing’ track was dedi-
cated to exploring the area of natural language processing, e.g., natural language
generation; dialog systems; speech-based applications; discourse, semantics, syntax,
grammar, and lexicon; lexical semantics, ontologies, etc. Six papers were accepted for
presentation in this track.
The ‘Modeling Human Cognitive Processes and Simulation’ track covered papers
on modeling perceptual processes; modeling of learning and thinking; modeling of
memory; brain–computer integration; collaborative learning systems; and computer
supported-assistive technology. Finally, four papers were selected for this track.
The ‘Image and Vision-Based Interactions’ track consisted of five papers covering
the area of interactive generation of textual and visual works; vision-based face,
motion, and gesture capture; scene understanding; vision-based medical imaging and
diagnosis; biometric interactive technologies; vision and languages; and vision for
robotics.
The ‘Applications of HCI’ track also comprised of five papers specifically repre-
senting the area of natural user interfaces; human-robot interaction; remote and
face-to-face collaboration; embodied conversational agents; mobile interfaces; tech-
nology for differently abled persons; health, multi-modal interfaces (speech, vision, eye
gaze, face, physiological information etc.) and intelligent visualization tools.
We would like to thank the Organizing Committee for their efforts and time spent to
ensure the success of the conference. We would also like to express our gratitude to the
Program Committee members for their timely and helpful reviews. And last but not
least, We thank all the invited speakers, authors, and members of the Program
VI Preface

Committee for their contribution in making IHCI 2018 a stimulating and productive
conference; and hope to get their continued support in the future.

December 2018 Uma Shanker Tiwary

 Organization

General Chair
Mriganka Sur MIT, USA

Steering Committee
Uma Shanker Tiwary IIIT Allahabad, India
Tom D. Gedeon Australian National University, Australia
Debasis Samanta IIT Kharagpur, India
Atanendu Sekhar Mandal CSIR-CEERI, Pilani, India
Tanveer Siddiqui University of Allahabad, India
Jaroslav Pokorny Charles University, Czech Republic
Sukhendu Das IIT Madras, India
Samit Bhattacharya IIT Guwahati, India

Program Committee
P. Nagabhushan IIIT Allahabad, India
A. G. Ramakrishnan Indian Institute of Science, India
Achard Catherine Sorbonne University, France
Aditya Nigam IIT Mandi, India
Akki B. Channappa IIIT Allahabad, India
Alexander Gelbukh Mexican Academy of Science, Mexico
Alexandre Vervisch-Picois Telecom SudParis, France
Alok Kanti Deb IIT Delhi, India
Amine Chellali Evry Val d’Essonne University, France
Amrita Basu JU, India
Anupam Agarwal IIIT Allahabad, India
Anupam Basu IIT Kharagpur, India
Ashutosh Mishra IIIT Allahabad, India
Atanendu Sekhar Mandal CSIR-CEERI, Pilani, India
Bernadette Dorizzi Telecom SudParis, France
Bibhas Ghoshal IIIT Allahabad, India
Catherine Achard Pierre et Marie Curie University, France
Daniel Wesierski Gdansk University of Technology, Poland
Daoudi Mohamed IMT Lille Douai, France
David Antonio Gomez ESTIA, France
Jauregui
David Griol Barres Carlos III University of Madrid, Spain
Debasis Samanta IIT Kharagpur, India
Dijana Petrovska Telecom SudParis, France
VIII Organization

Ekram Khan AMU, India

G. C. Nandi IIIT Allahabad, India
Geehyuk Lee KAIST, South Korea
Geoffrey Vaquette Commissariat à l’énergie atomique et aux énergies
alternatives, France
Gérard Chollet CNRS, France
Jan Platoš VŠB-TU Ostrava, Czech Republic
Jaroslav Pokorny Charles University, Czech Republic
Jérôme Boudy Telecom SudParis, France
José Marques Soares Universidade Federal do Ceará, Brazil
K. P. Singh IIIT Allahabad, India
Kavita Vemury IIIT Hyderabad, India
Keith Cheverst Lancaster University, UK
Laurence Devillers LIMSI, University ParisSud, Paris Saclay University,
France
Lopamudra Choudhury Jadavpur University, Kolkata, India
Malik Mallem Evry Val d’Essonne University, France
Manish Kumar IIIT Allahabad, India
Maria Stylianou Korsnes University of Oslo, Norway
Marion Morel Pierre et Marie Curie University, France
Martin A. Giese CIN/HIH University Clinic Tübingen, Germany
Michele Gouiffes LIMSI, ParisSud University, Paris Saclay University,
France
Mohamed Chetouani Pierre et Marie Curie University, France
Mriganka Sur MIT, USA
Nesma Houmani Telecom SudParis, France
Partha Pratim Das IIT KGP, India
Patrick Horain Telecom SudParis, France
Plaban Kumar Bhowmick IIT KGP, India
Pradipta Biswas CPDM-IISC, India
Pritish K. Varadwaj IIIT Allahabad, India
Harish Karnick IIT Kanpur, India
Somnath Biswas IIT Kanpur, India
Rahul Banerjee BITS Pilani, India
Richard Chebeir Pau et des Pays de l’Adour University, France
Samit Bhattacharya IIT Guwahati, India
San Murugesan BRITE Professional Services, Australia
Santanu Chaudhury CSIR-CEERI, Pilani, India
Satish K. Singh IIIT Allahabad, India
Shekhar Verma IIIT Allahabad, India
Shen Fang Pierre et Marie Curie University, France
Sukhendu Das IIT Madras, India
Suneel Yadav IIIT Allahabad, India
Tanveer J. Siddiqui University of Allahabad, India
Thierry Chaminade Institut de Neurosciences de la Timone, France
Tom D. Gedeon Australian National University, Australia
 Organization IX

Uma Shanker Tiwary IIIT Allahabad, India

Vijay K. Chaurasiya IIIT Allahabad, India
Vijayshri Tiwary IIIT Allahabad, India
Ydalia Garcia Telecom SudParis, France

Additional Reviewers
Ramesh Kumar Agrawal JNU, India
Sukumar Nandi IITG, India
Karmeshu JNU, India
Asif Ekbal IITP, India
Sriparna Saha IITP, India
Jaidhar C. D. NITK, India
Pinaki Mitra IITG, India
Sowmya Kamath S. NITK, India
Tirthankar Gayen JNU, India
Rajiv Singh Banasthali Vidyapith, India
Shiv Ram Dubey IIITS, India
Jean-Yves Didier Evry Val d’Essonne University, France
Mohammed Javed IIITA, India
Triloki Pant IIITA, India

Local Organizing Committee

Uma Shanker Tiwary IIITA, India
Shekhar Verma IIITA, India
Akki B. Channappa IIITA, India
Pritish K. Varadwaj IIITA, India
Vijayshri Tiwary IIITA, India
Ashutosh Mishra IIITA, India
Satish K. Singh IIITA, India
K. P. Singh IIITA, India
Vijay K. Chaurasiya IIITA, India
Bibhas Ghoshal IIITA, India
Manish Kumar IIITA, India
Suneel Yadav IIITA, India
Ajay Tiwary IIITA, India
Punit Singh IIITA, India
Rohit Mishra IIITA, India
Sudhakar Mishra IIITA, India
Shrikant Malviya IIITA, India
Santosh K. Baranwal IIITA, India
Pankaj Tyagi IIITA, India
 Contents

ECG, EEG-Based and Other Multimodal Interactions

Single Trial P300 Classification Using Convolutional LSTM

and Deep Learning Ensembles Method . . . . . . . . . . . . . . . . . . . . . . . . . . . 3
Raviraj Joshi, Purvi Goel, Mriganka Sur, and Hema A. Murthy

An Orthonormalized Partial Least Squares Based Spatial Filter

for SSVEP Extraction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 16
G. R. Kiran Kumar and M. Ramasubba Reddy

A Common Spatial Pattern Approach for Classification of Mental Counting

and Motor Execution EEG . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 26
Purvi Goel, Raviraj Joshi, Mriganka Sur, and Hema A. Murthy

A Real Time Human Emotion Recognition System Using Respiration

Parameters and ECG . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 36
C. M. Naveen Kumar and G. Shivakumar

Analysis of Action Oriented Effects on Perceptual Process of Object

Recognition Using Physiological Responses . . . . . . . . . . . . . . . . . . . . . . . . 46
Shanu Sharma, Anju Mishra, Sanjay Kumar, Priya Ranjan,
and Amit Ujlayan

Classification of EEG Signals for Cognitive Load Estimation Using Deep

Learning Architectures . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 59
Anushri Saha, Vikash Minz, Sanjith Bonela, S. R. Sreeja,
Ritwika Chowdhury, and Debasis Samanta

EEG Signals for Measuring Cognitive Development: A Study of EEG

Signals Challenges and Prospects . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 69
Swati Aggarwal, Prakriti Bansal, and Sameer Garg

EEG-Based Detection of Brisk Walking Motor Imagery Using Feature

Transformation Techniques. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 78
Batala Sandhya and Manjunatha Mahadevappa

Natural Language, Speech and Dialogue Processing

Cumulative Impulse Strength Based Epoch Extraction from Singing Voice. . . 93

Prigish George Abraham, M. S. Sinith, and A. R. Jayan
XII Contents

Chaos Analysis of Speech Imagery of IPA Vowels . . . . . . . . . . . . . . . . . . . 101

Debdeep Sikdar, Rinku Roy, and Manjunatha Mahadevappa

Generation of GMM Weights by Dirichlet Distribution and Model Selection

Using Information Criterion for Malayalam Speech Recognition . . . . . . . . . . 111
Lekshmi Krishna Ramachandran and Sherly Elizabeth

Social Choice Theory Based Domain Specific Hindi Stop Words List
Construction and Its Application in Text Mining. . . . . . . . . . . . . . . . . . . . . 123
Ruby Rani and D. K. Lobiyal

Effect of Devanagari Font Type in Reading Comprehension: An Eye

Tracking Study . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 136
Chetan Ralekar, Punyajoy Saha, Tapan K. Gandhi,
and Santanu Chaudhury

Bone Conducted Speech Signal Enhancement Using LPC and MFCC . . . . . . 148
Premjeet Singh, Manoj Kumar Mukul, and Rajkishore Prasad

Modeling Human Cognitive Processes and Simulation

Measuring Conceptual Incongruity from Text-Based Annotations . . . . . . . . . 161

Nisheeth Srivastava

Interactive Landslide Simulator: Role of Contextual Feedback in Learning

Against Landslide Risks. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 170
Pratik Chaturvedi and Varun Dutt

An Adaptive Cognitive Temporal-Causal Network Model of a Mindfulness

Therapy Based on Music . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 180
S. Sahand Mohammadi Ziabari and Jan Treur

Perceptual Filling-in of Blind-Spot for Surrounding Color

Gradient Stimuli . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 194
Amrita Mukherjee, Avijit Paul, Rajarshi Roy, Shibsankar Roy,
and Kuntal Ghosh

Image and Vision Based Interactions

Convolutional Neural Network Based Meitei Mayek Handwritten

Character Recognition . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 207
Deena Hijam and Sarat Saharia

Automated Fall Detection Using Computer Vision . . . . . . . . . . . . . . . . . . . 220

Pramod Kumar Soni and Ayesha Choudhary
 Contents XIII

Wild Animal Detection from Highly Cluttered Forest Images Using Deep
Residual Networks . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 230
Anamika Dhillon and Gyanendra K. Verma

A Non-deterministic Approach to Mitigate Replay Attack and Database

Attack Simultaneously on Iris Recognition System . . . . . . . . . . . . . . . . . . . 239
Richa Gupta and Priti Sehgal

Liveness Detection in Finger Vein Imaging Device Using

Plethysmographic Signals. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 251
Arya Krishnan, Tony Thomas, Gayathri R. Nayar,
and Sarath Sasilekha Mohan

Applications of HCI

Gesture Aided Voice for Voiceless . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 263

Dipti Jadhav, Tejaswini Koilakuntla, Sonam M. Mutalik Desai,
and Ramya Ramakrishnan

Towards Evaluating Architectural Design of Ancient Pilgrimage Site Using

Agent Based Modelling and Simulation . . . . . . . . . . . . . . . . . . . . . . . . . . . 272
Abha Trivedi and Mayank Pandey

Content-Based Music Classification Using Ensemble of Classifiers . . . . . . . . 285

Manikanta Durga Srinivas Anisetty, Gagan K Shetty,
Srinidhi Hiriyannaiah, Siddesh Gaddadevara Matt, K. G. Srinivasa,
and Anita Kanavalli

A Decision Tree Based Context-Aware Recommender System . . . . . . . . . . . 293

Sonal Linda and K. K. Bharadwaj

Mining of Influencers in Signed Social Networks: A Memetic Approach . . . . 306

Nancy Girdhar and K. K. Bharadwaj

Author Index . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 317

ECG, EEG-Based and Other
Multimodal Interactions
 Single Trial P300 Classification Using
Convolutional LSTM and Deep Learning
Ensembles Method

Raviraj Joshi1(B) , Purvi Goel1 , Mriganka Sur2 , and Hema A. Murthy1

1
Indian Institute of Technology Madras, Chennai, India
{rjoshi,purvi,hema}@cse.iitm.ac.in
2
Department of Brain and Cognitive Sciences,
Massachusetts Institute of Technology, Cambridge, USA
msur@mit.edu

Abstract. The odd ball paradigm is a commonly used approach to

develop Brain Computer Interfaces (BCIs). EEG signals have shown to
elicit a positive deflection known as the P300 event related potential dur-
ing odd ball experiments. BCIs based on these experiments rely on detec-
tion of the P300 potential. EEG signals are noisy, and therefore P300
detection is performed on an average of multiple trials, thus making them
inappropriate for BCI applications. We propose a neural network model
based on Convolutional Long Short Term Memory (ConvLSTM) for sin-
gle trial P300 classification. EEG data encodes both spatial and temporal
information using multiple EEG sensors. Convolutional neural networks
(CNNs) have been known to capture spatial information whereas LSTMs
are known to capture temporal information. Our experiments show that
the proposed method outperforms previous CNN based approaches on
raw EEG signals. The approaches were evaluated on publicly available
dataset II of BCI competition III. Another dataset was recorded locally
using audio beeps as stimuli to validate these approaches. The ensemble
models based on CNNs and ConvLSTM are also proposed. These models
perform better than individual architectures.

Keywords: Brain-computer interface · Event related potential

P300 · Convolutional neural networks · Convolutional long short term
memory

1 Introduction
Brain computer interfaces have made it possible to communicate our thoughts
without vocalizing or performing actions. It relies on the neural activity of the
brain to interface with the external world. BCIs are useful for the speech chal-
lenged since they enable such persons with a medium of communication. Elec-
troencephalography (EEG) is a widely used technique to develop BCI applica-
tions. EEGs are non invasive and provides a comparatively low cost solution to
build BCI devices.
c Springer Nature Switzerland AG 2018
U. S. Tiwary (Ed.): IHCI 2018, LNCS 11278, pp. 3–15, 2018.
https://doi.org/10.1007/978-3-030-04021-5_1
4 R. Joshi et al.

EEGs have a high temporal resolution as compared to other brain imaging

techniques. It is therefore capable of capturing very short cognitive processes in
the brain, which is elicited in the form of an Event Related Potential (ERP).
ERPs are the positive or negative potentials seen in the EEG signals in response
to external meaningful stimuli. These potentials are time locked and can be
reliably detected to build practical BCI. However, EEG signals have low signal
to noise ratio, which makes it difficult to detect these potentials in individual
trials. Traditionally, multiple trials are averaged together in the time domain so
that noise gets canceled and ERPs can be visualized. The focus of this work is the
ERP elicited during odd ball experiment known as P300 event related potential
and its single trial detection. In the odd ball experiment, a series of stimuli
are presented to the subject. Two types of stimuli are presented, one which is
frequent, while another that is infrequent. Infrequent stimuli cause ERP.
P300 signals have been mainly studied in the context of clinical diagnostics
and psychoanalysis. Recently P300 is being considered as a possible BCI. A very
popular program based on P300 is the P300 speller program [10]. P300 speller
is a tool used to efficiently communicate symbols of interest, by arranging them
in a 6 × 6 grid. There are other examples of P300 being used for cursor control
[14] and wheelchair control [21].
Detecting these P300 potentials in single trials has gained a lot of atten-
tion. The speed of communication will greatly improve if P300 can be reliably
detected from individual trials. Previous works have used conventional machine
learning techniques and neural network based classifiers for P300 classification
tasks. EEG data has both spatial and temporal characteristics. The spatial struc-
ture is attributed to different EEG sensors placed on the subject’s scalp and
the temporal structure is implicit as it picks up the voltage value at each time
instant. Moreover, P300 has timing characteristics and has different signature
across the spatial structure of the brain. Thus, the spatio-temporal information
is associated with P300 potential as well. In this work, we propose a neural
network model based on Convolutional Long Short Term Memory (ConvLSTM)
that naturally exploits both spatial and temporal characteristics [24]. As P300
potential is seen around 300 ms after the onset of stimulus for a short duration,
an LSTM based architecture was chosen to model this sequential information.
The convolution units of ConvLSTM encodes the spatial information captured
by electrodes placed on the scalp. This is the first work to employ ConvLSTM for
EEG data analysis. ConvLSTM also requires relatively less number of param-
eters. Thus, ConvLSTM is easy to train given the scarcity of training data in
EEG experiments.
We propose an architecture that uses a 3D representation of EEG data [6].
Although ConvLSTM has fewer parameters, it performs better than CNN archi-
tectures as it provides better modeling of EEG data. Finally, we evaluate the
performance of ensemble architectures based on the individual models discussed
in this paper. These ensemble models perform better than individual models.
The results of these experiments are better than the previous benchmarks on
publicly available BCI competition III dataset.
 Single Trial P300 Classification Using Convolutional LSTM 5

The rest of the paper is organized as follows: Sect. 2 describes related work
done in this field. The experimental setup, pre-processing methods and the mod-
els used for P300 detection are explained in Sect. 3. The results are discussed in
Sect. 4 and Sect. 5 is dedicated to Conclusion.

2 Related Work
Convolutional neural network (CNN) based approach for P300 detection was
first proposed by Cecotti and Graser [7]. The proposed model with two convo-
lutional and two fully connected layers could achieve comparable P300 detec-
tion results on raw EEG signals. The two convolutional layers were designed in
such a way that the first layer captured the spatial information and the second
layer captured the temporal information. They proposed seven CNN architec-
tures including three ensemble models. This work was extended by Liu et al. to
demonstrate the importance of batch normalization in the classification process
[15]. They proposed BN3 which coupled the existing CNN models with batch
normalization to prevent overfitting. BN3 could achieve the state of the art char-
acter recognition rates. In these works, the EEG data was represented in a 2D
matrix of dimension channels × time, thus forming an image like structure. The
convolution operations were performed along the spatial and temporal axis. A
more relevant EEG data representation was proposed by Carabez et al. [6]. In
their work, a single trial EEG was represented in 3 dimensions. The first two
dimensions encode the spatial structure of EEG while the third dimension is
the temporal axis. The work described in this paper uses the same technique to
represent the EEG signals.
Recently, Bashivan et al. in their work used stacked convolutional and LSTM
layers to process EEG data [5]. However, their work was based on spectral fea-
tures extracted in theta, alpha, and beta frequency bands. For each time instant,
EEG data was represented in (r,g,b) image like structure. The first two dimen-
sions encoded the spatial structure of channels while the third dimension was
reserved for the three frequency bands. Similar approach for EEG data repre-
sentation was employed by Maddula et al. for classifying P300 signals [16]. They
proposed different stacked CNN and LSTM architectures based on the same
spectral features. Their results were evaluated on P300 speller experiment which
was different from BCI competition III experiment and hence the results cannot
be compared. Our work is different from these works as we do not use stacked
architectures and our work is completely based on processing raw EEG signals.
Various other studies of using recurrent neural networks on EEG data have been
proposed over time [11,18,19], but these works fail to take advantage of spatial
characteristics of EEG data.
An ensemble of multiple machine learning models has been commonly used in
the classification of EEG signals [23]. Previously, to account for signal variability,
an ensemble of 17 support vector machines was used by Rakotomamonjy and
Guigue [20]. They also proposed a recursive channel elimination algorithm to
select discriminative channels. Deep learning ensemble models based on CNNs
were proposed in [4,7].
6 R. Joshi et al.

3 Methodology
Two different datasets are used in this work, one of which was collected in the
EEG Lab set up at Computer Science and Engineering Department, IIT Madras
and other was publicly available BCI Competition dataset [13].

3.1 BCI Competition Dataset

The dataset II from BCI competition III is used in our experiments. The data
consists of P300 evoked potentials generated using the P300 Speller paradigm
proposed by Farwell and Dochin [10].
The data was collected from 2 subjects in 3 sessions using a 64 channel EEG
sensor net. The sampling rate was 240 Hz and the signals were bandpass filtered
between 0.1 to 60 Hz. The odd ball paradigm results in unbalanced data for
the two classes, so the target P300 trials are replicated four times so as to get
an equal number of trials for both the classes. This is a classical oversampling
approach to counter data imbalance problem [17]. The number of samples for
both the classes during training and testing is shown in Table 1. A detailed
description of the dataset can be found in [13].

Table 1. Training and testing trials

Train Test
Subject Target Non-Target Target Non-Target
A 2550 12750 3000 15000
B 2550 12750 3000 15000

3.2 Odd Beep Experiment

The EEG data was collected using the Geodesic Sensor net comprising of 128
channels [2]. Subjects were students from different department in an age group
of 20 to 30 years. A written consent was given by participants before the exper-
iment. The subjects were seated in a comfortable chair to avoid fatigue inside
an acoustically and electromagnetically shielded testing chamber. The subjects
were asked to keep the eyes closed and do minimal muscle movement during the
recording.
The subjects are presented with a series of audio beeps with two different
frequencies. One of the beep also known as odd beep or target beep is presented
less frequently than the other non target beeps. The frequencies for two beeps
were chosen to be 1000 Hz (non target tone) and 2000 Hz (target tone). The
participant was supposed to gently tap on the mobile screen when a rare target
stimulus is presented and the other stimulus did not require any response. A train
 Single Trial P300 Classification Using Convolutional LSTM 7

of 150 beeps is presented to the subject. These 150 beeps are divided across 25
trials. Each trial, therefore, consists of 6 beeps. Out of these 6 beeps, 1 of the
beep is the target beep, while the others are non-target beeps. The target beep
is chosen randomly. Four extra non-target beeps were added at the start of the
experiment to make the subject familiar with non-target tone. The inter-stimulus
gap is set to 800 ms. A baseline of 60 s is recorded at the start and end of the
experiment. The experiment timeline is shown in Fig. 1.

Fig. 1. Single Oddbeep experiment timeline

The performance of a neural network is defined by the availability of enough

training samples. So multiple sessions of the experiment were performed on two
subjects. For each subject recording was done on 5 consecutive days and 3 ses-
sions were recorded on each day. A total of 15 sessions were recorded out of
which 10 sessions were used for training and 5 sessions were used for testing.
Table 2 shows the count of the target and non-target samples used for training
and testing. Target trials were replicated to match the number of non-target
trials. Although the number of samples is considerably less as compared to the
online dataset, it is just enough to train our shallow neural networks.

Table 2. Training and testing trials for Single Odd Beep Dataset

Train Test
Subject Target Non-Target Target Non-Target
A 250 1290 125 645
B 250 1290 125 645

3.3 Pre-processing
BCI Competition Dataset. The EEG data is highly dominated by noise, so
the data is bandpass filtered in the range 0.1–20 Hz. Moreover, the P300 wave is
characterized by a low frequency component which is captured well by the filtered
signal. The P300 potential is seen in EEG signals as a positive deflection around
300 ms after the onset of the stimulus, so we consider a 667 ms time window
after the onset of the stimulus [15]. As the data is sampled at 240 Hz, 667 ms
8 R. Joshi et al.

corresponds to 160 samples for each trial. The data segments are downsampled
and the length of segment is reduced to 80 samples. Thus, each target and non-
target trial is represented by a 64 × 80 matrix, where 64 defines the number of
channels.
In order to retain the spatial correlation of the electrodes, the 64 electrodes
are mapped to a 10 × 11 2D map as defined in [6]. Thus, each time instant is
represented by a 2D matrix and each trial is now 3 dimensional. Each column
vector of the 2D matrix in the original data corresponds to a 2D frame in the 3D
map. The depth of the 3D map represents the time axis (80 in this case). This
3D representation, similar to video frames is given as an input to our models.

Single Odd Beep Dataset. The pre-processing steps used for single odd beep
dataset are similar to the steps used for BCI competition dataset. Two major
changes are induced by the fact that this data is sampled at 250 Hz and is
captured using a 128 channel EEG net. The data is band passed filtered in the
range 0.1–20 Hz and a window of 640 ms after the onset of stimulus is extracted
for each trial. As the sampling rate is 250 Hz, 640 ms corresponds to 160 samples.
To preserve the spatial structure, the 128 electrodes are mapped to a 14 × 13
2D map. Some peripheral electrodes are dropped to create a dense matrix of size
10×11 as shown in Fig. 2. The 3-dimensional structure of electrodes is converted
into a 2D map using software provided by EEGLAB [9].

(a) 128 channel EGI Sensor net (b) Corresponding 2D Mapping

Fig. 2. Mapping of 128 electrodes to 2D matrix

4 Model Description
4.1 CNN Model for 3D Data (CNN-3D)
Convolutional neural networks have been widely used in classification tasks
involving images and videos. They are good at capturing spatial information
 Single Trial P300 Classification Using Convolutional LSTM 9

80 8 8 8

10
Output
Conv-2 Average 8
Conv-1
Pool
9
11 11 11 Sigmoid
Input
128
FC Layer

Fig. 3. Model 1: Architecture of CNN-3D

and can be used to process 3D data. CNNs are especially suited to EEG because
of their translation invariance property. The EEG electrodes may be slightly mis-
placed across recordings. This shift of electrodes is easily accounted by convolu-
tion and pooling layers. Our model uses average pooling instead of the standard
max pooling based on the fact that nearby electrodes have similar signatures.
The model architecture is shown in Fig. 3 and the details of the proposed model
are listed below

– The input data is batch normalized and passed to the next layer. The impor-
tance of batch normalization for EEG data is already studied in [15]. The
batch size is set to 64 samples.
– The data of size 10 × 11 × 80 is passed through a convolutional layer. This
layer has a filter of size 3 × 3 which spans across the depth of input. Eight
such filters are applied and appropriate padding is done to retain the input
dimension. After this operation, the output is reduced to 10 × 11 × 8 which is
passed through relu non-linearity. This output is further batch normalized and
a dropout of 0.2 probability is applied. The dropout operation is performed
only during the training phase.
– The next layer performs a similar convolution operation with the same num-
ber of filters.
– The output is then average pooled with a filter of size 2 × 2 and a stride of 1.
This reduces the output dimension to 8 × 9 × 8. This is followed by a batch
normalization layer and the 3D output is reshaped to a 1D vector.
– The flattened input is passed through a dense layer with 128 neurons and
sigmoid activation. This is followed by a dropout of 0.2 probability.
– Since this is a binary classification problem, the final layer is a dense layer
with one neuron and sigmoid activation. Binary cross entropy is used as a
loss function with Adam optimizer [12]. Learning rate is kept at 0.001.

4.2 Convolutional Long Short Term Memory (ConvLSTM)

ConvLSTMs are similar to LSTMs with the dense operations replaced by convo-
lution operations. As convolution operations use parameter sharing, the number
of parameters is greatly reduced in ConvLSTM. Moreover, ConvLSTM allows
10 R. Joshi et al.

Fig. 4. Model 2: Architecture of ConvLstm

simultaneous learning of spatio-temporal features. Therefore, ConvLSTM is pre-

ferred over stacked convolution and LSTM architecture [22]. The 3D EEG data
is passed as a sequence of 2D frames to the ConvLSTM. This is equivalent to
passing a series of images to ConvLSTM. Although CNNs are capable of cap-
turing temporal information, the number of parameters increases exponentially
as the depth of input increases. This makes it difficult to capture long term
relationships. We, therefore, explore the usage of ConvLSTM to model spatial
and temporal structure of EEG. The placement of electrodes on the scalp, the
mental state of the subject across sessions and trials can lead to variability. This
is accounted by the convolutional operations of ConvLSTM. The temporal char-
acteristics of EEG are captured by the LSTM structure. The model architecture
is shown in Fig. 4 and the details of the proposed model is described below

– The input data is first batch normalized and passed to the next layer. The
batch size is set to 64 samples.
– The sequence of inputs in the form of 2D map is passed through recurrent
convolutions of filter size 3 × 3. Three such filters are used and the output
is passed through tanh non-linearity. A dropout of 0.2 is used along with
recurrent dropout of 0.1. This is followed by a batch normalization layer.
– A similar recurrent convolution is again applied on the output of the previous
layer followed by batch normalization.
– The sequence returned is flattened and a dropout of 0.5 is applied. The final
layer is a dense layer having only one neuron with sigmoid activation function.
Binary cross entropy loss is used along with Adam optimizer for training.
Learning rate is kept at 0.001.
 Single Trial P300 Classification Using Convolutional LSTM 11

4.3 Ensemble Models

The individual models proposed in this paper along with BN3 proposed in [15]
were used to build ensemble classifiers. The individual classifiers were trained
on the entire dataset and best models were picked using 10% validation data
for ensemble testing. Scores from individual models were fused using simple
averaging. Four ensemble models were evaluated using combinations of CNN-
3D, BN3, and ConvLSTM.

5 Results and Discussion

The performance of single trial P300 detection is measured for BN3, CNN-3D,
ConvLSTM and their ensembles. Several metrics that are used to evaluate the
performance of the models are true positive (TP), false positive (FP), true neg-
ative (TN), false negative (FN) and others mentioned below

TP TP
Precision = Recall =
TP + FP TP + FN
TP + TN Precision.Recall
Recognition = F-measure = 2
TP + TN + FP + FN Precision + Recall

The number of trainable parameters for each of the models under considera-
tion is shown in Table 3. Along with CNN-3D and ConvLSTM, the BN3 model
is used in our analysis for comparison of the results [15]. ConvLSTM model has
one-third parameters as compared to BN3 and CNN-3D. This allows the model
to learn representations even with less data. These models were implemented
using Keras library [8] with tensorflow backend [3]. AWS p2.xlarge instances
were used for evaluation of models [1]. The results of BN3 were obtained under
this environment, original results can be seen in [15]. Ensemble models use indi-
vidual models having similar performance.

Table 3. Number of model parameters

BN3 CNN-3D ConvLSTM

39,649 32,593 11,213

5.1 Analysis of BCI Competition Dataset

The complete results are shown in Table 4. Models CNN-1, MCNN-1 and MCNN-
3 were proposed by Cecotti and Graser [7]. Their results are listed here for com-
parative analysis. CNN-3D gives better performance than the existing models
as it takes into consideration the spatial positioning of electrodes. It classifies
12 R. Joshi et al.

Table 4. Performance of different models for Online Dataset.

Data Model TP TN FP FN Recog. Recall Precision F1-score

Set
III A BN3 + 1926 11,748 3252 1074 0.7597 0.6420 0.3720 0.4710
CNN3D +
CONVLSTM

CNN3D + 1963 11,592 3408 1037 0.7531 0.6543 0.3655 0.4690

CONVLSTM

BN3 + 1919 11,701 3299 1081 0.7567 0.6397 0.3678 0.4670

CONVLSTM

BN3 + 1952 11,573 3427 1048 0.7513 0.6507 0.3629 0.4659

CNN3D

CONVLSTM 1928 11,582 3418 1072 0.7505 0.6426 0.3606 0.4620

CNN3D 1938 11,536 3464 1062 0.7486 0.6460 0.3588 0.4613

∗
BN3 1910 11,229 3771 1090 0.7299 0.6367 0.3362 0.4400
CNN-1 2021 10,645 4355 979 0.7037 0.6737 0.3170 0.4311
MCNN-1 2071 10,348 4652 929 0.6899 0.6903 0.3080 0.4260
MCNN-3 2023 10,645 4355 977 0.7038 0.6743 0.3172 0.4314
III B BN3 + 2003 12,810 2190 997 0.8229 0.6677 0.4777 0.5569
CNN3D +
CONVLSTM

CNN3D + 2115 12,371 2629 885 0.8048 0.7050 0.4458 0.5462

CONVLSTM

BN3 + 1956 12,864 2136 1044 0.8233 0.6520 0.4780 0.5516

CONVLSTM

BN3 + 1990 12,762 2238 1010 0.8196 0.6633 0.4707 0.5506

CNN3D

CONVLSTM 2099 12,330 2670 901 0.8016 0.6997 0.4401 0.5404

CNN3D 2139 12,099 2901 861 0.7910 0.7130 0.4244 0.5321

BN3∗ 2009 12,329 2671 991 0.7966 0.6697 0.4293 0.5232
CNN-1 2035 12,039 2961 965 0.7819 0.6783 0.4073 0.5090
MCNN-1 2202 11,453 3547 798 0.6899 0.7340 0.3830 0.5034
MCNN-3 2077 11,997 3003 923 0.7038 0.6923 0.4089 0.5141

the examples of both the classes more accurately than the existing models. The
proposed CNN-3D model has a high recognition rate as well as high precision
and recall.
Results show that ConvLSTM models the spatio-temporal information well
and performs better than other architectures. Table 4 shows that CNN-3D per-
forms well for the target class whereas ConvLSTM performs better for the non-
target class. Therefore, in order to retain the properties of both the models,
 Single Trial P300 Classification Using Convolutional LSTM 13

Table 5. Performance of different models for Single Odd Beep Dataset.

Data Model TP TN FP FN Recog. Recall Precision F1-score

Set
A BN3 + CNN3D + CONVLSTM 88 603 42 37 0.8974 0.7040 0.6769 0.6902
CNN3D + CONVLSTM 90 599 46 35 0.8948 0.7200 0.6618 0.6897
BN3 + CONVLSTM 84 599 46 41 0.8870 0.6720 0.6462 0.6588
BN3 + CNN3D 88 604 41 37 0.8987 0.7040 0.6822 0.6929
CONVLSTM 82 596 49 43 0.8805 0.6560 0.6259 0.6406
CNN3D 91 595 50 34 0.8909 0.7280 0.6454 0.6842
BN3 84 588 57 41 0.8727 0.6720 0.5957 0.6316
B BN3 + CNN3D + CONVLSTM 74 606 39 51 0.8831 0.5920 0.6549 0.6218
CNN3D + CONVLSTM 74 592 53 51 0.8649 0.5920 0.5827 0.5873
BN3 + CONVLSTM 75 607 38 50 0.8857 0.6000 0.6637 0.6302
BN3 + CNN3D 77 598 47 48 0.8766 0.6160 0.6210 0.6185
CONVLSTM 72 590 55 53 0.8597 0.5760 0.5669 0.5714
CNN3D 78 573 72 47 0.8454 0.6240 0.5200 0.5673
BN3 74 572 73 51 0.8389 0.5920 0.5034 0.5441

we propose ensemble models consisting of the existing BN3 architecture, the

proposed CNN-3D, and ConvLSTM architectures.
The ensemble models consisting of BN3 + CNN-3D architectures and BN3
+ ConvLSTM architectures have a very high recognition rate since they classify
the non-target class examples with a very high precision whereas the model
consisting of CNN-3D + ConvLSTM architectures have a very high recall.

5.2 Analysis of Single Odd Beep Dataset

Table 5 shows the results for different models. The results are similar to previ-
ous analysis, CNN-3D and ConvLSTM based models perform better than BN3.
However, the recognition rates for odd beep dataset is better than BCI compe-
tition dataset. This can be due to the better spatial resolution provided by 128
channel EEG net. Although its scalability to more number of trials still needs
to be validated.

6 Conclusions

The results obtained in this work show that the 3D input for single trial P300
classification works better than the 2D input. The reason can attributed to
the fact that 3D input capture spatial surroundings of the data very well. CNNs
based on 3D input captures the information well but may lack the temporal infor-
mation. Spatio-temporal characteristics of EEG are captured well by ConvL-
STM, and hence give better results using less parameters. Ensemble approaches
that take into account the properties of both models perform significantly better.
14 R. Joshi et al.

References
1. Amazon ec2 - p2 instances. https://aws.amazon.com/ec2/instance-types/p2/.
Accessed 09 Jan 2018
2. The geodesic sensor net. https://www.egi.com/research-division/geodesic-sensor-
net. Accessed 9 Jan 2018
3. Abadi, M., et al.: Tensorflow: a system for large-scale machine learning. OSDI 16,
265–283 (2016)
4. Barsim, K.S., Zheng, W., Yang, B.: Ensemble learning to EEG-based brain com-
puter interfaces with applications on P300-spellers
5. Bashivan, P., Rish, I., Yeasin, M., Codella, N.: Learning representations from
EEG with deep recurrent-convolutional neural networks (2015). arXiv preprint.
arXiv:1511.06448
6. Carabez, E., Sugi, M., Nambu, I., Wada, Y.: Convolutional neural networks with
3D input for P300 identification in auditory brain-computer interfaces. Comput.
Intell. Neurosci. 2017 (2017)
7. Cecotti, H., Graser, A.: Convolutional neural networks for P300 detection with
application to brain-computer interfaces. IEEE Trans. Pattern Anal. Mach. Intell.
33(3), 433–445 (2011)
8. Chollet, F., et al.: Keras (2015). https://github.com/fchollet/keras
9. Delorme, A., Makeig, S.: Eeglab: an open source toolbox for analysis of single-trial
eeg dynamics including independent component analysis. J. Neurosci. Methods
134(1), 9–21 (2004)
10. Farwell, L.A., Donchin, E.: Talking off the top of your head: toward a mental
prosthesis utilizing event-related brain potentials. Electroencephalogr. Clin. Neu-
rophysiol. 70(6), 510–523 (1988)
11. Fedjaev, J.: Decoding EEG brain signals using recurrent neural networks (2017)
12. Kingma, D.P., Ba, J.: Adam: a method for stochastic optimization (2014). arXiv
preprint arXiv:1412.6980
13. Krusienski, D., Schalk, G.: Wadsworth BCI dataset (P300 evoked potentials), BCI
competition III challenge (2004). http://www.bbci.de/competition/iii/
14. Li, Y., Long, J., Yu, T., Yu, Z., Wang, C., Zhang, H., Guan, C.: An EEG-based BCI
system for 2-D cursor control by combining mu/beta rhythm and P300 potential.
IEEE Trans. Biomed. Eng. 57(10), 2495–2505 (2010)
15. Liu, M., Wu, W., Gu, Z., Yu, Z., Qi, F., Li, Y.: Deep learning based on batch
normalization for P300 signal detection. Neurocomputing 275, 288–297 (2018)
16. Maddula, R., Stivers, J., Mousavi, M., Ravindran, S., de Sa, V.: Deep recurrent
convolutional neural networks for classifying P300 BCI signals. In: Proceedings of
the Graz BCI Conference (2017)
17. Mazurowski, M.A., Habas, P.A., Zurada, J.M., Lo, J.Y., Baker, J.A., Tourassi,
G.D.: Training neural network classifiers for medical decision making: the effects
of imbalanced datasets on classification performance. Neural Netw. 21(2–3), 427–
436 (2008)
18. Naderi, M.A., Mahdavi-Nasab, H.: Analysis and classification of EEG signals using
spectral analysis and recurrent neural networks. In: 17th Iranian Conference of
Biomedical Engineering (ICBME), pp. 1–4. IEEE (2010)
19. Petrosian, A., Prokhorov, D., Homan, R., Dasheiff, R., Wunsch II, D.: Recurrent
neural network based prediction of epileptic seizures in intra-and extracranial EEG.
Neurocomputing 30(1–4), 201–218 (2000)
 Single Trial P300 Classification Using Convolutional LSTM 15

20. Rakotomamonjy, A., Guigue, V.: BCI competition III: dataset II-ensemble of SVMs
for BCI P300 speller. IEEE Trans. Biomed. Eng. 55(3), 1147–1154 (2008)
21. Rebsamen, B., Guan, C., Zhang, H., Wang, C., Teo, C., Ang, M.H., Burdet, E.:
A brain controlled wheelchair to navigate in familiar environments. IEEE Trans.
Neural Syst. Rehabil. Eng. 18(6), 590–598 (2010)
22. Sainath, T.N., Vinyals, O., Senior, A., Sak, H.: Convolutional, long short-term
memory, fully connected deep neural networks. In: IEEE International Conference
on Acoustics, Speech and Signal Processing (ICASSP), pp. 4580–4584. IEEE (2015)
23. Sun, S., Zhang, C., Zhang, D.: An experimental evaluation of ensemble methods
for eeg signal classification. Pattern Recognit. Lett. 28(15), 2157–2163 (2007)
24. Xingjian, S., Chen, Z., Wang, H., Yeung, D.Y., Wong, W.K., Woo, W.c.: Convolu-
tional LSTM network: a machine learning approach for precipitation nowcasting.
In: Advances in neural information processing systems, pp. 802–810 (2015)
 An Orthonormalized Partial Least
Squares Based Spatial Filter for SSVEP
Extraction

G. R. Kiran Kumar(B) and M. Ramasubba Reddy

Indian Institute of Technology Madras, Chennai 600036, India

kirankumar.g.r@hotmail.com, rsreddy@iitm.ac.in

Abstract. In this study, a novel orthonormalized partial least squares

(OPLS) spatial filter is proposed for the extraction of the steady-state
visual evoked potential (SSVEP) components buried in the electroen-
cephalogram (EEG) data. The proposed method avoids over-fitting of
the EEG data to the ideal SSVEP reference signals by reducing the over-
emphasis of the target (pure sine-cosine) space. The paper presents the
comparison of the detection accuracy of the proposed method with other
existing spatial filters and discusses the shortcomings of these algorithms.
The OPLS was tested across ten healthy subjects and its classification
performance was examined. Further, statistical tests were performed to
show the significant improvements in obtained detection accuracies. The
result shows that the OPLS provides a significant improvement in detec-
tion accuracy across subjects compared to spatial filters under compar-
ison. Hence, OPLS would act as a reliable and efficient spatial filter
for separation of SSVEP components in brain-computer interface (BCI)
applications.

Keywords: Steady-state visual evoked potential (SSVEP)

Electroencephalogram (EEG) · Brain-computer interface (BCI)
Orthonormalized partial least squares (OPLS)

1 Introduction

Steady-state visual evoked potentials (SSVEP) are electroencephalogram (EEG)

components that are generated over the visual cortex in response to periodically
flicking visual stimuli. They are elicited in response to flicker frequencies greater
than 4 Hz [9] and the SSVEP amplitude is modulated by visual spatial attention
provided by the user [11]. Further, the SSVEP response is in-phase with the
target frequencies and contains other harmonics. Due to its properties, relatively
high signal to noise ratio (SNR) and ease of implementation, SSVEP has been
studied increasingly for application in non-invasive brain-computer interfaces
(BCI) [6].

This work was supported by Indian Institute of Technology Madras.

c Springer Nature Switzerland AG 2018
U. S. Tiwary (Ed.): IHCI 2018, LNCS 11278, pp. 16–25, 2018.
https://doi.org/10.1007/978-3-030-04021-5_2
 An OPLS Based Spatial Filer for SSVEP Extraction 17

SSVEP based brain-computer interfaces requires very low training and

exhibits high information transfer rates (ITR) compared to other BCI modalities
[13]. A classical SSVEP-BCI system consists of flickering target stimuli placed
at different locations. The user selects the target by gazing over it and intended
target is extracted by analysing the acquired EEG for components correspond-
ing to the flicker frequency [12]. Extracting low noise SSVEP components from
a given EEG data segment is a basic and crucial step in SSVEP detection meth-
ods. The band limiting of the acquired data is useful in eliminating noise outside
the desired SSVEP frequency range. But, the filtering does not remove the noise
embedded in the desired range. Hence, to achieve higher SSVEP SNR, a number
of spatial filtering techniques employing linear signal models have been proposed.
Common spatial filtering techniques used in SSVEP based BCIs include
hardware-based methods like best bipolar combination (BCC) and multivari-
ate data analysis (MVA) algorithms like principal component analysis (PCA),
minimum energy combination (MEC), maximum contrast combination (MCC),
and partial least squares (PLS) spatial filter [7]. Even though BCC provides
considerable improvement in SNR, the selection of the optimal electrode pair
need to be done through an exhaustive search for every individual user. The
PCA is an unsupervised method that exploits the common information between
the input EEG channels to maximize the variance of the reconstructed data and
disregards the SSVEP source model [4].
Other MVA methods are supervised linear signal models that use a sim-
ple SSVEP model consisting of sine-cosine signals as target data for improving
the EEG signal components. The MEC and MCC spatial filters try to min-
imise the SSVEP noise or maximise SNR by computing the signal and noise
components using ordinary least squares (OLS) regression method. The OLS
has several shortcomings as it fails when the inter-channel correlation increases
(multicollinearity) and it further assumes that the target SSVEP model as fixed
[3]. Recently, a PLS spatial filter has been proposed that overcomes the disadvan-
tages of MEC and MCC by efficiently dealing with highly correlated channels.
The PLS tries to maximise the covariance between the EEG and SSVEP model
and considers the target SSVEP model to contain error, thus providing a robust
estimate. Among the reported spatial filters in the literature, MCC and PLS
have been shown to achieve the highest performance [8].
In this study, we propose an orthonormalized partial least squares (OPLS)
spatial filter that rewards the channels that better model the information con-
tained in the features of the SSVEP target space. In PLS spatial filter, the input
and output spaces with very high variance are overemphasized even if the cor-
relation between them in the projected data is not significant. The OPLS is
a variant of PLS which overcomes this disadvantage by minimizing the mean
squares error (MSE) instead of the covariance and not considering the variance
of the target (pure sine-cosine) space. The proposed method is evaluated by
comparing its SSVEP detection performance in terms of accuracy with classical
SSVEP spatial filters such as PLS and MCC using EEG data collected from
ten subjects. Further, statistical tests are performed to depict the improvement
achieved using OPLS method.
Another random document with
no related content on Scribd:
 CHAPTER XVIII
PRINCESS VICTORIA EUGENIE OF BATTENBERG AS QUEEN OF SPAIN

1906

As the Spanish authoress Concepción Gimeno de Flaquer devotes

the last chapter of her book, “Mujeres de Regia Estirpe” (Women of
Royal Degree), to Queen Victoria Eugénie of Spain, it seems that I
should fall short of the mark were I not to publish some of the
Spanish impressions of the present English Queen at the Court of
Spain.
 V I C TO R I A E U G É N I E , Q U E E N O F S PA I N

Señora Flaquer says: “The presence of the beautiful Princess at

the royal palace is like a shining star on a dark night, a soft balmy
breath of wind in a violent storm, a refreshing dew in hot weather,
and a ray of hope in depression.”
This description is Spanish in its imagery, and it is interesting to
note the more measured language in which Figuerola Ferretti
expresses the joy of Spain at the news of the engagement:
 “The news is like a fresh spring of hope to us Spaniards, who
regard any English girl as a symbol of sincerity and sweetness, and
how much more so when that girl is grand-daughter of the great
Queen Victoria, whose name is venerated throughout the Peninsula!
“Whilst regarding the entry of Princess Ena into Spanish spheres
as the commencement of a new era for the education and progress
of our women, who are only waiting for the opportunity to prove their
intellectual worth, I must say I might have some fears lest the
Princess should be chilled by the restrictions of Spanish Court
etiquette, had not King Alfonso already shown himself capable of
breaking down the unnecessary barriers which would prevent his
future bride continuing the happy outdoor life and the social
pleasures which brighten the existence of royal ladies in England.
“‘Manners maketh man,’ it is said, but it is also true that ‘man
maketh manners,’ and when our monarch follows the natural and
noble impulses of his heart, it is always to the making of a manner
which expresses good feeling.
“The young Spaniard has marked with great interest King
Alfonso’s foreign mode of courtship, which oversteps the lines of our
customs; and as he follows in the footsteps of the royal fiancé, he
will soon see that invigorating motor-car excursions and walks in a
garden with the queen of his heart are more conducive to mutual
knowledge of character than perpetually thrumming on a guitar
outside the lady’s window, or only being permitted to whisper words
of love in a corner of a room where the rest of the family is
assembled.
“To judge from ancient records, the arrival of the young Princess
Eleanor of England in 1170 as the bride of Alfonso VIII. of Spain led
to a reaction against the strictures of etiquette introduced by the
Moors to the extreme limitations of the liberty of our ladies; and it
was by the natural assumption of a certain freedom of action that the
daughter of young Henry II. passed a happy life of nearly half a
century as Queen-Consort in our country. And Princess Ena is not
likely to fall short of her English predecessor in her natural love of
liberty.
“Readers of Mariana’s ‘History of Spain’ may be struck with the
resemblance of the meeting of the young royal lovers on the borders
of Spain in 1170 and that of the illustrious couple at Biarritz. The
ardent young Alfonso VIII. was charmed with his English Eleanor,
even as our Alfonso XIII. admired the Ena of your land; and as
Queen Eleanor associated herself with the promotion of learning and
letters for men, and supported the foundation of the University of
Palencia, our future Queen Ena will doubtless encourage the present
movement for the education of girls, which has just culminated in the
opening of the Middle-Class College under the committee of ladies
of the Ibero-American Society, presided over by Queen Maria
Cristina.”
The joy foretold by the Spanish courtier was more than realized
at the arrival of the English bride. Her bright, sunny smile and ready
acknowledgment of the people’s evident admiration of their future
Queen delighted the people.
But the tragedy of the bomb cast in the bouquet, which caused so
much disaster, came like a sudden frost, and nipped the
spontaneous joy of the young Queen, and the drives and walks in
the city of Madrid became a source of fear instead of joy. It is hard to
us here in England to realize what the bomb outrage on her
marriage-day was to Queen Victoria of Spain.
 Q U E E N V I C TO R I A O F S PA I N A N D T H E Q U E E N - M O T H E R AT A
BULL-FIGHT

Wearers of the Victoria Cross and the D.S.O. have not often gone
through such a terrible ordeal. For soldiers on active service are at
least prepared for such tragedies, but in the glitter and gaiety of a
marriage-day the blow was dealt in the dark.
An officer in the Wad Ras Regiment, who was close to the
carriage, told me that he can hardly bear to speak of it even now.
The gaily-decorated street was suddenly transformed into the fearful
scene of a battle-field. The cries of the dying and the sight of the
killed sent many people out of their minds. With the calm courage of
a soldier’s daughter, Queen Victoria neither swooned nor went into
hysterics; but the shock went deep into her soul, and she naturally
fears a repetition of the horror when she is in the city.
The people, therefore, are a little disappointed at their greetings
not meeting with the quick response of the first days in her new land;
and as Spaniards would do anything for a smile, and love to see
happiness, this inborn terror, begotten of the tragedy of her wedding-
morn, would form a barrier between the English Queen and her
people, were they not reminded of the source of the set expression
on her face.
In La Granja this is different. The freedom of the country life gives
scope again for our Princess’s smiles, and the beautiful gardens and
the charm of the palace seem far removed from the tragedy of the
city.
“Oh, how we adore her when she is like that!” said the simple-
hearted, sympathetic Spaniards, as they saw the eager, guileless
way the Queen showed her young cousin, Princess Beatrice of
Coburg, her lovely country residence; and after she had passed up
the fine staircase of the palace, lined by the halberdiers sounding
their drum tattoos of welcome, she appeared at one of the windows
to smile on the soldiers as they saluted her in their parade past the
palace.
 P R I N C E F E R D I N A N D O F B AVA R I A , B R O T H E R - I N - L AW O F
ALFONSO XIII.

Bouquets are naturally, of course, still looked upon with suspicion

at the Spanish Court. When Miss Janotha, the celebrated pianist,
wished to leave a beautiful bouquet at the palace as an offering to
Princess Henry of Battenberg, when she was in Madrid, the lackey
looked at it askance, saying:
“We are not to take bouquets.”
 Miss Janotha looked regretful, and I was very glad when a
superior official stepped forward and said:
“We do not take bouquets, but as it comes with the English lady
we know here, it is all right.”
This confidence I acknowledged gratefully; the Polish pianist was
pleased, and the bouquet was taken.
“The Queen is always her bright, merry self on the yacht,” said a
distinguished naval officer, when speaking of the shock of the bomb
to the young Queen. “She sings, and is as happy as the day is long,
for there is no fear of such tragedies on board ship.”
One always connects Spain with sunshine, and Queen Victoria
was interested at seeing the after-effects of a snow-storm in Madrid.
Their Majesties sallied forth in a motor-car to the park of the Retiro.
The Queen expressed her admiration at the clever efforts in statuary
made of the snow which had fallen in the morning. The newly-
appointed Prime Minister, Maura, was easily distinguished as a
snow-man, and many other celebrities were recognized in this
exhibition of snow-figures made by the street gamins. Great lions in
front of the War Office also showed the skill of the officials in turning
the snow into form when clearing the pathways, and in the squares
and streets there were many presentments, both male and female.
The Infanta Maria Teresa was driving across the Puerta del Sol
with her young husband during the inclement weather, when a mule
of her carriage slipped on one of the tram-lines, which form a perfect
network at this busy centre, and the carriage came to a standstill.
The Princess descended from the vehicle, and would have walked
home had she not herself slipped on the treacherous footwalk.
Fortunately, the etiquette which formerly forbade a commoner to
touch royalty even in a time of danger does not now prevail, and a
policeman raised the Infanta from the ground, and placed her in a
tram, in which the rest of the journey to the palace was made.
 T H E I N FA N TA M A R I A T E R E S A , A F T E RWA R D S W I F E O F P R I N C E
F E R D I N A N D O F B AVA R I A

An Audience at the Court of Spain.

It was one morning during this short season of snow in 1907 that
a charming Spanish lady, Señora doña Carmen Burgos de Seguí,
called to ask if I would join her and two or three other members of
the well-known Andalusian Centre in their visit to the palace to invite
King Alfonso XIII. and Queen Victoria to a forthcoming fête to be
held by the Centre at a theatre. All the formalities with reference to
the audience had been arranged, and I was pleased to accept the
invitation to join the commission.
As a fall of snow precluded the possibility of being able to obtain
a carriage or cab—for the cobbled stones of the roads make it
unsafe for horses in slippery weather—I put on my snow-shoes and
fur cloak, and soon arrived with my companions at the royal palace,
which flanks the whole side of the great Plaza de Oriente, and
towers majestically above the richly-wooded valley of the River
Manzanares.
The white-cloaked sentries, in their three-cornered hats, saluted
us respectfully as we passed, and the colonnaded, rich-carpeted
staircase soon led us to the gallery which lines the quadrangle of the
royal palace.
A sympathetic porter helped me to remove my cloak and
overshoes, and as I shook out my dress and donned my white
gloves he said:
“Her Majesty will be very pleased to see a compatriot, for since
last June she has not seen an English lady.”
The ring of the halberds on the floor of the gallery as the historic
halberdiers changed guard, and the quick word of command, were
the only sounds to break the solemn silence as darkness fell on the
courtyard, where snow was falling softly.
A lackey in gold livery now issued from the royal apartments and
met us in the gallery. He then conducted us to an imposing doorway
leading to the landing of the state double-winged staircase, which is
only used on very important occasions. It was in this gallery that the
young King and his sister, Maria Teresa, startled Queen Victoria, on
her first Shrove Tuesday in Spain, by jumping out at her disguised
with masks.
 The white marble lions, the blazing lights of the fine chandeliers,
the rich carpets, the carved marble rails and handsome walls, looked
like a scene in a fairy-tale as we saw it for the first time, and after
passing several footmen and officials on the landing we reached an
antechamber, where we were asked to wait our turn of audience.
The walls of this salon were hung with rural scenes embroidered
on tapestry set in crimson velvet. Large mirrors reached from the
floor to the painted ceiling, and reflected the crystal candelabra and
the works of art which lined the room, with its crimson-satin-covered
furniture on a velvet-pile carpet.
Just before we were summoned to the royal presence, I was told
it was contrary to Court etiquette to wear a veil, so I removed it in
time to obey the summons of the Court official, who appeared with
the papers relating to our visit; and being handed over to the usher,
we ran the gauntlet of the eyes of Chamberlain and military men
standing about in uniform in every salon, and passed through a large
anteroom with green-satin-panelled walls hung with pictures of the
royal predecessors of the present King, and thence into a room like a
large and splendid ballroom, where a lady was sitting on duty in full
Court dress with a companion, and we were finally ushered into the
presence of the King and Queen.
The Queen looked fair and regal as she stood in the beautifully
decorated French salon in a perfectly-made pale pink dress trimmed
with the finest lace, and the King was in the undress uniform of a
Captain-General.
The Queen looked somewhat sad as she graciously received us,
and she must, indeed, have thought that it was another wearisome
occasion of speeches and remarks which would be in an unknown
tongue to her. According to the etiquette of the Spanish Court, the
King and Queen were both standing to receive us in the beautiful
little boudoir. Indeed, the room seemed only arranged for such
audiences.
My introduction to the King as an Englishwoman at once met with
a cordial shake of the hand and a pleasant “How do you do?” after
the Queen had gracefully greeted us. As Her Majesty looked pleased
to see somebody from her native land, I begged to be allowed to
address Her Majesty, and, passing behind the King to her side, I
soon had the great delight of hearing her speak with pleasure of the
Shakespearian Bazaar in London, where I had last seen her as
Princess Ena; and when speaking of my friend, Miss Janotha, she
said, “Yes, I have known her since I was so high,” holding her hand a
little distance from the floor.
In the talk with the Spanish ladies, Alfonso pleased one who has
rather advanced opinions by the gusto with which he said, “Yes,
there are indeed far too many associations in Spain!” for this remark
showed that His Majesty is alive to the evil; and if the clerical party
would only allow action to be taken to prevent this overwhelming
number of religious associations in Madrid, it would be to the joy of
the country.
For these associations ply their trades of printing, chocolate-
making, boot-making, needlework, etc., and they undersell the
trades of the lay-workers, as they have neither taxes nor rent to pay.
This abuse the Government was seeking to remove by bringing in a
law for the diminution of such societies, but the camarillas of the
palace, instigated by the clerical party, checked the progress which
Canalejas, the President of the Congress, was making in this
direction, by causing the fall of the Ministry. It was falsely reported at
the palace that Canalejas is atheistic and antagonistic to the Church,
whereas he told me himself that he is very religious. He has a private
chapel in his house, where Mass is celebrated every day. But, as the
Minister said, this matter of the associations (of which many are from
Belgium, France, and other parts of the Continent) militating against
the trade of Spain is a matter of State policy, and has nothing to do
with religion.
“And now the King is offended with me, and I have no chance of
an explanation with His Majesty,” said the ex-Minister, who a short
time before had been patted on the back for his zeal for the welfare
of the land.
When I looked at the young Queen, so tall, so elegant, and so
alone in a foreign land, I felt how difficult it must be to fulfil her rôle to
the satisfaction of all parties.
The report that the expected royal heir’s layette was to be made
entirely in Spain excited much commendation; but when I went to
see the things at the best shop in Madrid, I could but note that they
were not so fine as I had expected.
“No, no,” said the proprietor of the place; “all the best things are
made in the convents, and we have only the second and third best.
The Queen, I believe, meant to benefit the trade of Madrid, for she
was so sweet and gracious when she called here, but the priests
gave most of the work to the societies in which they are interested.”
Moreover, the King not only expressed himself frankly about the
associations at our audience at Court, but he showed a deep interest
in the details of the Andalusian fête to which we had come to invite
Their Majesties. It is the King’s keenness in all matters which
captivates those about him.
“What dances will there be?” he asked eagerly. “And will there be
songs of the Region?” he queried. To all these questions the Spanish
ladies answered, flattered at the interest manifested.
During the talk I was privileged to have with Her Majesty in
English, I was charmed with her evident affectionate recollection of
things in England, whilst graciously interested in the subjects which
had brought me to Spain.
 She smiled sweetly when I kissed her hand on leaving, as I said I
did not know whether I did it as a Spanish subject or as an English
compatriot, but in either case it was an honour I could not forego.
Then, the audience over, we were conducted with the same
pomp and ceremony as before through the stately salons and
guarded galleries till we were once more in the free atmosphere of
the Plaza de Oriente, environed by the statues of past Spanish
Sovereigns, who looked spectral in the moonlight, and met by editors
who wished to make copy out of our audience.
The King said we could see the state apartments of the palace on
the following day, but, as the weather was bad, I proved to be the
only one who appeared the next morning to profit by the royal
permission.
There was much discussion in the Chamberlain’s office as to the
right course to pursue about my visit. The royal permission, which is
rarely granted whilst Their Majesties are in residence, had been
given to the party of ladies, and only one had come. Was that one to
be given the privilege or no? I was amused at hearing the flow of
oratory which the subject aroused among those in the office, but
directly I suggested myself deferring the visit to another day, the
traditional courtesy of the Spaniards gained the day, and with many
bows and protestations of pleasure I was escorted past the sentries
on guard by a courtly guide, who did the honours of the salons. If I
describe these state apartments in the words of Pierre Loti, it will be
seen that I do not exaggerate their magnificence, for the French
author writes:
“The place is decorated by Velasquez, Bayeu, Tiepolo, Mengs,
Luis Lopez, Rubens, Vicente Lopez, Luis Gonzalez, etc. A whole
world of splendour seems to unfold, and as one passes through what
seems an interminable line of salons, all marked with the particular
ideas of the artists employed on them, one is struck by a series of
surprises.
“The great frames of the doors are all made of agate or rare
marbles, whose colours and veined surface harmonize beautifully
with the brocades of the walls.
“The Salon of Charles III. is hung with blue satin starred with
silver. Other salons are hung with exquisite old satin, with furniture of
the time of Louis XV.; others are hung with an inimitable red
embroidered with gold of the time of the Renaissance, or with pale
green curiously blended with yellow or saffron colour, or deep blue
embossed with yellow, with the stiff but elegant furniture of the
Empire period.
“Then there is a salon with the whole ceiling and panels of
faience, and when the artist died before completing the work, his
wife finished it by inaugurating and superintending the exquisite
embroidery of garlands of white and pink roses on panels of grey
silk.”
There is another salon with the walls covered with cherubs of the
white pottery for which the factory of the Retiro was famous—viz.,
the throne-room, with its ceiling painted by Tiepolo, its crimson-satin-
hung walls, its long mirrors, its many crystal chandeliers, its busts of
the Roman Emperors on pedestals, and, above all, its magnificent
throne with its crimson and gilt chairs. The four steps of the throne
are guarded by two large lions of gilt brass, and the royal seats are
flanked by figures representing the cardinal virtues; and the
banqueting-hall, with its magnificent columns, panels of porphyry
and marble, is a perfect picture.
Spanish ladies declare that Victoria of Spain looked every inch a
Queen when she first took her seat by her royal Consort. Her
diadem-crowned golden hair, beautiful face, and her exquisite
toilettes, make a striking feature at the State receptions; and when
we consider that it was in an unknown tongue the talk went on, it
was wonderful she could preserve her stately and quiet demeanour.
Now the Queen has become mistress of the Spanish tongue, her
subjects can admire her intellectual as well as her physical charms.

The Court of Spain at Candlemas.

The protocol of the royal Court etiquette at Madrid and the rites of
the Roman Catholic Church produce a pageant in the Spanish
palace at the Feast of the Purification (commonly called Candlemas)
which, in splendour and solemnity, savour more of the Middle Ages
than of the present practical period.
The galleries on the first-floor of the magnificent quadrangular
Palace of Madrid showed the advent of a great event, for the
windows looking on to the spacious colonnaded courtyard were
hidden by the fine tapestries of the same character that lined the
walls on the opposite sides. Rich carpets covered the floors, and the
companies of stalwart halberdiers, the Guard of the palace, were
placed at ten o’clock along the corridor, bearing on their shoulders
their halberds with the inscription, “Fabrica de Toledo, Alfonso XIII.,
1902,” which were presented to them when the present King was
added to the list of the Sovereigns to whom the corps had the
honour to be the bodyguard. Officials of the palace and officers
constantly passed to and fro, giving orders and seeing that the
soldiers stood in their right places.
The three-cornered hats edged with white, the high black
leggings reaching to the white breeches, and the blue coat
decorated with scarlet badges bearing the castle and the crowned
lion, is the same uniform of the Royal Guard as it was in the early
part of the last century, and it reminds one of the pictures of
Napoleon, etc., of that time.
 A clap of the hands from a Court official announced the opening
of a large door leading to the apartments of the Infanta Maria Teresa
and her husband, Prince Ferdinand of Bavaria. Bright and happy
looked the young Princess as she passed along, with her ready
sweet smile for familiar faces, and looking quite pretty in her pale
blue dress. The merry eyes of the stalwart, fair young Prince were
cast about in cheerful greetings as he swung along in his striking
blue and scarlet hussar uniform, with the jacket slung on one
shoulder, revealing the richly embroidered sleeves underneath.
There was a pause after the young couple passed to the seats
set apart for the Royal Family in the chapel; then the strains of a
march from an opera were heard from the band of the Royal
Halberdiers in the courtyard below, the halberdiers stood at attention,
and the royal procession was seen coming along the gallery.
The gentlemen of the Court, with the badges marking their
respective offices, the Chamberlain, all in full dress, with white silk
stockings and richly embroidered coats, were followed by the
grandees and officers in their striking uniforms. They walked in two
single files, so as to leave clear the view of the Royal Family. The
Infantas of Bavaria and the Infanta Isabel came with their respective
Ladies and Gentlemen in Waiting in full Court dress. The widowed
Prince of Asturias was in his place, and lastly came the King in his
uniform as Admiral, and wearing the Order of the Golden Fleece and
the Collar of Carlos III., and the procession solemnly passed through
the guarded portals of the chapel, where the Queen-mother and the
young Queen Victoria had already taken their places. For after
December 25, 1886, when a special service was held in the royal
chapel of the palace, in which the Virgin’s protection was petitioned
for the young Sovereign, the Court was in gala costume for two
days. A reception was held, congratulations received, and from then
till the birth of the expected heir Queen Victoria did not sit with the
King on the throne in the chapel, but in the royal box on the ground-
floor. All eyes were soon turned in admiration to the youthful English
Sovereign of Spain, who looked like a beautiful picture in her white
mantilla shading her diamond-crowned beautiful hair, and dressed in
a rich, soft white Court dress.
The doors of the chapel were soon again flung open, the
halberdiers were again called to attention, and the procession issued
from the chapel in the same order in which it had entered, only now it
was preceded by the Canons of the palace and other clerics in
gorgeous vestments, with the Archbishop of Sion in gold-and-white
mitre and emblazoned cope; and everybody in the procession
carried a long candle, as they solemnly made the tour of the gallery
to the tune of the psalm of old in which Simeon declared that the
Babe brought to the Temple would be “a Light to lighten the
Gentiles.”
The King, as he bore his candle, looked ruefully at his sister, as
much as to say: “How am I to manage this?” The Infanta smiled
pleasantly, and her young husband’s eyes twinkled with fun. The
evident strain on the dignity of the stately grandees and
Chamberlains to carry their lights befittingly gave a touch of humour
to the stateliness of the scene, and I overheard a grandee say, when
he was asked by one behind him not to walk so slowly: “I can’t go
any quicker, or I shall spill some grease on the Infanta’s train!”
The tour of the galleries made, the procession returned to the
chapel, the King went back to his throne, and Queen Victoria of
Spain to the royal box, the Chamberlains, grandees, Court ladies,
the Infanta Maria Teresa, the Infante Ferdinand, Don Carlos, and the
Infantas of Spain, all knelt reverently with their candles, whilst the
incense was swung in front of the King after he had partaken of the
Holy Sacrament.
Then, when the candles were removed by the Chamberlains, the
strains from the beautiful stringed orchestra accompanied the fine
voices of the hidden choir, which swelled in harmony in the chants of
the occasion. The lofty cupola of the chapel, with its mythical
painting supported by the gilt cherubs poised above the marble and
porphyry-pillared panels of the walls, were a fitting setting to the
scene.
Then the candles were once more handed round, and the
glittering company again knelt in prayer. When the torches were
finally taken from the worshippers, the assembly all left the chapel in
solemn order, each grandee kneeling in turn for a second before the
altar, and crossing himself before saluting the Queens in the royal
box. The Infanta Maria Teresa, the Infanta Isabel, and the Court
ladies, made a low reverence to both the Queens in the royal box
before leaving the chapel, and the King, with his characteristic
freedom from the fetters of etiquette, disregarded the scarlet mat,
and knelt on the carpeted floor for a minute before the altar; and then
with his natural grace he made a respectful salute to both his mother
and his wife, and left the church, to pass once more with his retinue,
and followed by the military, along the tapestry-lined galleries to the
royal apartments.
The Court of Spain is especially noted for its cult of symbolism.
The events of the Church calendar are presented in a realistic way
which is suggestive of the Middle Ages.
I believe the Courts of Spain and Austria are alone in their
dramatic representation of Christ’s act of washing the feet of the
disciples and feeding them on the eve of the Crucifixion.

How the King washes the Feet of the Beggars

and feeds them on Maunday Thursday.
 It is only by special invitation from the chief Court Chamberlain
that one can witness the King’s performance of this religious function
on Maunday Thursday. Being the fortunate possessor of this permit, I
passed at three o’clock in the afternoon to the Hall of Columns in the
palace. There the Court soon assembles in state, the ladies in
magnificent dresses, of which the trains are tastefully arranged by
the Gentlemen-in-Waiting over the backs of the chairs behind them,
and the throng of nobles, Ministers, and officers in their gorgeous
uniforms, make a brilliant show.
The King soon appears, attended by the Bishop of Alcalá and
Sion, some clerics, and twelve grandees in Court attire. After
divesting himself of his sword, Alfonso is girded with a towel by the
prelate, and passes to the line of beggars, who sit humbly waiting for
the honour which is to be paid them.
These poor men are chosen by lottery about a fortnight preceding
the function, and their feet naturally undergo a course of preparation
prior to the ceremony, and they are all swathed in the long Spanish
cloaks given them for the occasion.
The twelve grandees in attendance have meanwhile knelt in front
of the twelve beggars and taken off their shoes, and the forms of
these stately personages in this humble position make a sort of
screen between the eye of the public and the King’s action of
passing a towel over the feet of the poor men, which have been
sprinkled from the gold ewer of the Bishop who precedes the
Sovereign.
The King then passes to the long table, of the form and laid in the
style familiar to us in pictures of the Last Supper, and the beggars
are handed by their respective grandees to their seats at the board.
The poor men on the last occasion were blind, but this in no way
affected their calm acceptance of the fact of being the cynosure of a
Court in splendid state and the object of their Sovereign’s service.
Stolid were the faces as the King swiftly passed the items of the long
menu before their sightless eyes, and as the smell of the good things
was wafted to their nostrils they knew that time would give them a
more substantial realization of the dainties.
For the dish of each part of the menu found its way to the baskets
for the respective beggars, after being handed by the King to the
grandees in attendance. Thus twelve large pieces of salmon, twelve
joints of beef, and a dozen dishes of every item, were distributed by
the august purveyor.
The menu finished, His Majesty completed the programme by
handing also the glasses and cruets to the distinguished retinue,
they also finding their way to the poor guests; and finally the King
concluded the function by folding up the tablecloth with the zest
characteristic of his actions.
The final privilege granted to these beggars on Maunday
Thursday is the sight of the state apartments. This benefit seems to
be thrown away on those whose affliction deprives them of the
appreciation of their splendour, but etiquette must be preserved.

On Good Friday the King exercises his power of pardoning

criminals, so he stands in front of the high-altar, and, raising to
heaven the gold salver containing the names of the privileged
persons, he says: “These I pardon for their crimes, even as I hope
God will pardon my sins.”
The carving of the lamb on Easter Sunday is quite a religious
function at the King’s table. The Bishop of Sion has a service of
benediction, and the King and Queen take their places in state on
this occasion.
One of the most striking ceremonies preceding the birth of a royal
infant in the palace is that of transporting the arm of St. John the
Baptist, a sash said to have belonged to the Virgin Mary, and other
relics, from the chapel to the bedroom of the Queen. The King and
the Court all take part in the function, attended with all the ceremony
due to the occasion, and so fatiguing is the ritual that in May, 1907,
Queen Victoria nearly fainted during the performance. Indeed, so
many are the wearisome rites which Queen Victoria had to follow,
according to the customs of the Court of Spain, that more than one
editor of a democratic paper declared that if he were interested in the
royal succession he would see that the authorities did not thus
imperil it.
On Saturday afternoon the King and Queen go to hear the Salve
in a quiet, simple fashion at the Church of the Buen Suceso. Women
who press their hungry children to their bosoms as they gaze up into
the face of the young Queen as she sits in the royal box on this
occasion wonder if Her Majesty knows what their sufferings are. The
rise in the price of bread, which the Spanish Press speaks of as an
act of unjustifiable oppression, recently drove the women to
desperation, and made them break the windows of the bakers’ shops
in some quarters of the city. This strong measure was successful,
and bread is now at its usual price; for, as a Spanish lady said, “The
determination of hardly-driven mothers can accomplish more than
the discussions of men.”
The poor people who greeted the Queen with such loud
acclamations on her arrival in Spain wonder, moreover, if she knows
that the liberal gifts bestowed on such festivals as the King’s Saint’s
Day (January 23) to the orphans of the Sacred Heart of Jesus, the
Real Associación de Beneficencía Domiciliaría, etc., are devoted to
the maintenance of the friars and nuns of these associations rather
than to the benefit of the needy.
The Queen’s philanthropic spirit is, moreover, only appealed to
on behalf of the orphanages and schools in the hands of the clerics,
and so she is not in touch with the lay side of her country’s efforts.
 “If such serious matters as the lessening of the heavy duties on
articles of food which go to the support of the friars, and the limitation
of the associations which kill our industries, are not soon settled by
the Government, they will be settled in the street!” say many
thoughtful men in Spain; and it was those who saw the seriousness
of the aspect who expressed their disappointment that the English
Queen was so gracious in her reception of the deputation which
presented the King with a petition, signed by leading ladies of
fashion, against the Law of Associations; for these ladies are
naturally unable to realize the struggles of their sisters against the
monopoly by these associations of many of the industries on which
their bread depends, such as chocolate-making, perfume-distilling,
embroidery, lace-making, etc.
A bitter smile wreathes the lips of people as they read of the royal
sympathy for these organizations, but they say: “What can one
expect, when the young Queen is only environed with Spanish
ladies, whose support of the clerics smoothes their lives, and with
the Spanish priests, who dictate every deed of sympathy to the
Sovereigns of Spain?”
In speaking of Queen Victoria, it must be remembered that all
opinions expressed are modified by the reminder of the Queen’s
difficulty of knowing the real circumstances of a strange land of
which she had to master the language, and that conventional
greetings, gala receptions, and State dinners, do not lead to a true
knowledge of the country and its needs.
It is hoped by patriots that the Queen’s advent will lead to the
adoption of a system of Parliamentary elections in Spain similar to
that of England; for, as everybody says, if the deputies of the
Congress were elected by the votes of the people instead of by the
voice of the Ministers, the country’s conventional love would be
cemented into real devotion to the dynasty, and the reforms would
be enacted which would save the land from stagnation and poverty.
 The article published in an English review by a Spaniard, called
“Spain’s Hopes of a New Era,” showed that the English Queen was
looked upon as the coming saviour of the country from much that
has so far crippled it; and the twenty-four short articles and poems
published in the Woman’s Agricultural Times from the pens of
leading literary and professional ladies, begging their future
Sovereign to encourage the lighter branches of agriculture as
professions for women, show the hoped-for result of the new reign.
Disappointment has been expressed that this spontaneous act
from Spanish women of note, many of whom have influence in the
Press, has not so far resulted in any royal act of encouragement in
the direction desired; so the people do not know whether their
Sovereign is in sympathy with their needs or not. Directly Her
Majesty is in touch with the more progressive women of her country
the Press will be filled with the fact, and the warm hearts of the
people will beat with gratitude, and they will be able to talk about
more than the beauty of the Queen’s hair and complexion.
The Spanish Court seemed to surpass itself in magnificence in
the splendid functions of the christening of the first heir to the throne.
Every seat in the chapel of the royal palace was reserved for
those of the highest degree, and the gallery along which the royal
procession passed on its way from the royal apartments to the
church was crowded by people, who could only gain admittance by
tickets from the Chief Chamberlain of the palace.
The magnificent tapestries only used on State occasions were
displayed, the halberdiers lined the way, and the ladies, all in
mantillas, with their cavaliers in uniform or evening dress, waited in
breathless impatience for the advent of the new Prince of Asturias.
At last came the announcing hand-clap, and with solemn, stately
step the procession came round the angle of the gallery.
 CHRISTENING OF THE PRINCE OF ASTURIAS, SON OF
ALFONSO XIII.

First came the mace-bearers, then the ushers, all in double file,
then two long lines of Chamberlains in gold-laced coats and white
silk stockings, followed by the grandees of Spain in their striking
military uniforms and feathered cocked hats. Then came seven
grandees carrying the seven salvers with such requisites for the holy
ceremony as a salt-cellar, a gold basin and ewer, a cut lemon, a lace
towel, a cape, and a large cake. Behind this party came the royal
Prince himself, looking really an ideal infant in his beautiful laces. His
fair little uncovered head, and sweet, placid, tiny face, and clenched
fists were the admiration of all beholders. He was in the arms of the
Marquesa de los Llanos, who is the chief of his retinue, and on one
side walked the Nuncio, who is the representative of His Holiness, as
godfather, and on the other was the Queen-mother, as the
godmother. The King looked dignified in his new position, as father.
The Infantes and Infantas followed, with their suites. The Infanta
Maria Teresa and her husband, Infante Fernando, being only
convalescent from measles, were unable to be present. Don Carlos,
the widowed husband of the King’s late sister, the Infanta Mercedes,
led little Prince Alfonso, who was known as the heir to the throne
until the birth of his little cousin, and by the way he tripped along and
evidently enjoyed the brilliant sight he seemed in no way saddened
by his deposition from his former rank.
It was then understood that Don Carlos would marry before long
the beautiful daughter of the Princess of Orleans.
The little sister of the ex-heir was led by the hand by the Infanta
Isabel, at whose side walked Princess Henry of Battenberg,
beautifully robed in grey velvet and ermine. Prince Arthur of
Connaught, with Captain Wyndham and the Princes from Russia and
Germany, etc., all had their places in the procession. China was also
represented. The personal staff of the King was conspicuous, and
the halberdier band of music marshalled the glittering throng to the
chapel.
The altar was decorated with white flowers. The historic font in
which the members of the Royal Family have for centuries been
baptized was in the centre of the chapel.
Thirty-six Bishops and four Cardinals officiated. The royal
neophyte was very good in the arms of his grandmother, Queen
Maria Cristina. The water sprinkled on his brow was from the River
Jordan. The christening ceremony over, the King decorated his infant
son with the Order of the Golden Fleece, the Order of Isabella the
Catholic, and the Collar of Charles III. All the ladies of the Court were
in full dress.
Then the procession filed back to the royal apartments in the
same order in which it had come. The dresses of the ladies of the
nobility were all rich in colour and profuse with splendid jewels. The
white satin, gold-embroidered train of the Duchess of Arion set off
the beauty of her person.
 Amid the many stately personages, the majestic figure of Sir
Maurice de Bunsen was conspicuous, and Lady de Bunsen attracted
attention by her beauty and her beautiful and yet simple Court dress.
The ceremony was, indeed, one not easily to be forgotten as the
occasion of a gathering of important personages or their
representatives from far and near, and no infant could have taken its
prominent part on such an important occasion with greater
equanimity than did the beautiful babe—the Prince of Asturias. To
sleep and to smile seem easy things to do, but to do them during the
solemn, stately functions in which Church and State meet together to
do him honour is not always an easy thing for an eight-day-old infant,
and by accomplishing this task little Prince Alfonso added to the
affection and admiration with which he is regarded.
It is always pleasant to Queen Victoria Eugénie to pass from the
pageantry and pomp of the palace of Madrid to the less formal
surroundings of the country. It is by no mere figure of speech that it
can be said that when they are at the Palace of San Ildefonso, at La
Granja, King Alfonso and Queen Victoria Eugénie lead the simple
life.
The King rises early in the morning, and takes a long walk or ride
—sometimes alone, sometimes attended by one gentleman, and
sometimes accompanied by the Queen—or he has a bicycle spin in
the grounds.
Not long since, when the King was driving alone with the Queen
in a motor, he saw a soldier thrown from his horse, upon which he
immediately jumped from his automobile and rushed to the
assistance of the poor man.
The King’s interest in his soldiers is very marked, and when the
bell rang at dinner-time, when he was in consultation one morning
with the commanding officer, he went with him to inspect the food,
and tasted it himself.
 The Prime Minister could hardly hide his surprise, when he
arrived from Madrid one day to transact political business, to meet
his Sovereign in his shirt-sleeves, the young King having taken off
his coat, as it was a hot day for golfing.
King Alfonso and Queen Victoria like to go about unattended
together, and the Spanish custom of wearing no hat in the country
has been quite adopted by the English Sovereign; and people in the
little town are pleased to see the Queen pass by on foot to pay a visit
to some friends without anything on her head, but, of course,
carrying a parasol. Both the Sovereigns spend hours with their baby
son in the beautiful gardens of La Granja. The King will often take
him in his arms and carry him about, or if they meet the baby Prince
in his little white carriage when they are out walking they stop and
fondle and talk to him.
The Queen is beginning to share King Alfonso’s interest in golf,
and, indeed, she takes her part well in the game, and can easily do
the full round over the rough ground without any sign of fatigue.
When the weather is too hot for golf, Her Majesty much enjoys a
peaceful afternoon by the river, trout-fishing. In this sport she is quite
an expert, and the large basket of fish caught by the Queen and the
Duquesa de San Carlos was carried home in triumph on one
occasion, and figured on the royal menu for dinner.
In the Court, surrounded by courtiers and people, whose rôle is to
please, the Queen may hardly gauge the depth of Spain’s devotion
to their English Sovereign.
When the people of Galicia presented the island of Cortegada to
Alfonso XIII., they said it was also a tribute of sympathy to their
Queen. “It will be nice for Her Majesty to be within sight of the
English ships as they lie at anchor off the coast,” they said. “It will be
easy to go to England from there, and she will perhaps be reminded
of her Isle of Wight. Then, we hope to see King Edward in the
Spanish island home.”
The enthusiasm for England is very great since the royal alliance,
and for the successful recommendation of any fashion, game, or
sport it is only necessary to say it is English.
It is, therefore, hoped that a nearer acquaintance with our
Parliamentary system will lead to its adoption in Spain.
As, in face of his overwhelming influence, it is not possible for the
people to elect a deputy of either party in opposition to the one
chosen or supported by the cacique of the district, the deputies
elected by public vote have mostly been republicans. Hence the
suffrage is associated with republicanism in Spain, and Catalonia,
where this has been successful, is connected with the idea of
Separatism. Thus, with the misreport of things in Madrid, it is thought
that Catalonia is wanting in Monarchists. But whenever the wisdom
of the King leads to a royal visit to Barcelona, the enthusiasm for the
royal visitor always proves that the Press has misinterpreted the
state of feeling there; and the welcome that will be given to Queen
Victoria when she makes the long-looked-for visit to Barcelona will
show that Catalonia is also content that an English Queen should
reign over them.
At Cortegada the peasants to whom I was introduced made the
sign of the cross, for they said they had never seen an
Englishwoman before; but they had one for their Queen, and she
was welcome in the land.
“Viva la Reina Victoria!” was the cry which floated across the
moonlit waters as the peasants returned to the mainland after the
celebration of their annual festival on the island which had been
offered for the acceptance of the King and the Queen, and, indeed,
this cry is echoed throughout the land.
 INDEX

A
Alagon, Duke of, 79, 81, 84, 85, 121
Alava, General, 72, 73, 101
Alcalá Galiano, historian, 3
Alfonso XII.: birth, 193;
education, 198–202;
made King, 248;
character, 252–4, 280;
travels through Spain, 255;
marriage, 257;
second marriage, 264;
visit to Germany, 268;
return, 269–71;
ill-health, 271, 275;
escapades, 271–2;
government, 273;
death, 276
Alfonso XIII., 294–6, 303;
chapter on, 305–13;
baptism, 305;
instruction, 306–8;
youthful occupations, 308–9;
coronation oath, 309–10;
affection for his mother, 310, 315;
grants audience to authoress, 320;
personal traits, 322–3;
ceremony on Maunday Thursday, 332–4;
daily life, 341–2