Image and Video Technology 8th

Pacific Rim Symposium PSIVT 2017

Wuhan China November 20 24 2017
Revised Selected Papers 1st Edition
Manoranjan Paul
Visit to download the full and correct content document:
https://textbookfull.com/product/image-and-video-technology-8th-pacific-rim-symposiu
m-psivt-2017-wuhan-china-november-20-24-2017-revised-selected-papers-1st-edition
-manoranjan-paul/
More products digital (pdf, epub, mobi) instant
download maybe you interests ...

Image and Video Technology 9th Pacific Rim Symposium

PSIVT 2019 Sydney NSW Australia November 18 22 2019
Proceedings Chilwoo Lee

https://textbookfull.com/product/image-and-video-technology-9th-
pacific-rim-symposium-psivt-2019-sydney-nsw-australia-
november-18-22-2019-proceedings-chilwoo-lee/

Computational Intelligence and Intelligent Systems 9th

International Symposium ISICA 2017 Guangzhou China
November 18 19 2017 Revised Selected Papers Part II
Kangshun Li
https://textbookfull.com/product/computational-intelligence-and-
intelligent-systems-9th-international-symposium-
isica-2017-guangzhou-china-november-18-19-2017-revised-selected-
papers-part-ii-kangshun-li/

Advances in Multimedia Information Processing – PCM

2017: 18th Pacific-Rim Conference on Multimedia,
Harbin, China, September 28-29, 2017, Revised Selected
Papers, Part II Bing Zeng
https://textbookfull.com/product/advances-in-multimedia-
information-processing-pcm-2017-18th-pacific-rim-conference-on-
multimedia-harbin-china-september-28-29-2017-revised-selected-
papers-part-ii-bing-zeng/

Information Security and Cryptology: 13th International

Conference, Inscrypt 2017, Xi'an, China, November 3–5,
2017, Revised Selected Papers 1st Edition Xiaofeng Chen

https://textbookfull.com/product/information-security-and-
cryptology-13th-international-conference-inscrypt-2017-xian-
china-november-3-5-2017-revised-selected-papers-1st-edition-
Image and Graphics 9th International Conference ICIG
2017 Shanghai China September 13 15 2017 Revised
Selected Papers Part III 1st Edition Yao Zhao

https://textbookfull.com/product/image-and-graphics-9th-
international-conference-icig-2017-shanghai-china-
september-13-15-2017-revised-selected-papers-part-iii-1st-
edition-yao-zhao/

Music Technology with Swing 13th International

Symposium CMMR 2017 Matosinhos Portugal September 25 28
2017 Revised Selected Papers Mitsuko Aramaki

https://textbookfull.com/product/music-technology-with-
swing-13th-international-symposium-cmmr-2017-matosinhos-portugal-
september-25-28-2017-revised-selected-papers-mitsuko-aramaki/

Cryptology and Network Security 16th International

Conference CANS 2017 Hong Kong China November 30
December 2 2017 Revised Selected Papers Srdjan Capkun

https://textbookfull.com/product/cryptology-and-network-
security-16th-international-conference-cans-2017-hong-kong-china-
november-30-december-2-2017-revised-selected-papers-srdjan-
capkun/

Digital TV and Wireless Multimedia Communication: 14th

International Forum, IFTC 2017, Shanghai, China,
November 8-9, 2017, Revised Selected Papers 1st Edition
Guangtao Zhai
https://textbookfull.com/product/digital-tv-and-wireless-
multimedia-communication-14th-international-forum-
iftc-2017-shanghai-china-november-8-9-2017-revised-selected-
papers-1st-edition-guangtao-zhai/

Service Research and Innovation 5th and 6th

Australasian Symposium ASSRI 2015 and ASSRI 2017 Sydney
NSW Australia November 2 3 2015 and October 19 20 2017
Revised Selected Papers 1st Edition Amin Beheshti
https://textbookfull.com/product/service-research-and-
innovation-5th-and-6th-australasian-symposium-assri-2015-and-
assri-2017-sydney-nsw-australia-november-2-3-2015-and-
 Manoranjan Paul
Carlos Hitoshi
Qingming Huang (Eds.)
LNCS 10749

Image and
Video Technology
8th Pacific-Rim Symposium, PSIVT 2017
Wuhan, China, November 20–24, 2017
Revised Selected Papers

123
Lecture Notes in Computer Science 10749
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, 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/7412
Manoranjan Paul Carlos Hitoshi
•

Qingming Huang (Eds.)

Image and
Video Technology
8th Pacific-Rim Symposium, PSIVT 2017
Wuhan, China, November 20–24, 2017
Revised Selected Papers

123
Editors
Manoranjan Paul Qingming Huang
School of Computing and Mathematics University of Chinese Academy of Science
Charles Sturt University Beijing
Bathurst, NSW China
Australia
Carlos Hitoshi
University of São Paulo
São Paulo
Brazil

ISSN 0302-9743 ISSN 1611-3349 (electronic)

Lecture Notes in Computer Science
ISBN 978-3-319-75785-8 ISBN 978-3-319-75786-5 (eBook)
https://doi.org/10.1007/978-3-319-75786-5

Library of Congress Control Number: 2018934319

LNCS Sublibrary: SL6 – Image Processing, Computer Vision, Pattern Recognition, and Graphics

© Springer International Publishing AG, part of Springer Nature 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.

Cover illustration: Yellow Crane Pagoda, Wuhan. Photo by Reinhard Klette, Auckland, New Zealand

Printed on acid-free paper

This Springer imprint is published by the registered company Springer International

Publishing AG part of Springer Nature
The registered company address is: Gewerbestrasse 11, 6330 Cham, Switzerland
 Preface

The Pacific-Rim Symposium on Image and Video Technology (PSIVT) 2017 took
place in amazing Wuhan, China. Wuhan is a city that changes virtually every day: An
extensive subway system and a more than 30-km-long walk- and bicycle way around
the famous East Lake are just two of the very recent additions to Wuhan. Previ-
ous PSIVT symposiums were held in Taiwan, Chile, Japan, Singapore, South Korea,
Mexico, and New Zealand; all host countries sharing a coastline with the Pacific Ocean.
The 2017 issue of PSIVT had four keynote speakers: Xiaoyi Jiang from the
University of Münster, Germany, with his talk entitled “Biomedical Imaging: A
Galisonian Perspective for Sciences,” Shuicheng Yan from the National University of
Singapore about “Deep Learning Towards On-Device Visual Analytics,” Martin
Constable from the Royal Melbourne Institute of the Arts, Vietnam, about “Color
Contrast in the Aesthetic Image: An Examination of the Complex Ways That Color
Contrast Manifests in Paintings,” and Domingo Mery from the Pontificia Universidad
Católica, Chile, with a talk entitled “Modern Computer Vision Techniques for X-ray
Testing in Baggage Inspection.” We thank all four keynote speakers for presenting
wonderful talks that were informative, sometimes surprising, and altogether very much
stimulating for those working in the addressed fields.
PSIVT 2017 attracted 197 submissions to the main conference and the five work-
shops taking place the day before the main conference. This volume contains all the
accepted submissions presented at the main conference. Workshop papers are pub-
lished in a separate LNCS volume. Each submission to the main conference was
reviewed in full by at least three (and up to four) reviewers, before area chairs made
recommendations and the three program co-chairs the final decision. In the main
conference, 19 papers were accepted for oral presentation and 20 papers were accepted
for poster presentation. The acceptance rate is 21% for oral presentations and 43%
overall for the main conference. According to reviewing results we selected the three
top-ranked accepted papers for the first oral session of the conference. An especially
formed Award Committee decided for these three outstanding papers:
The best paper award went to Tao Rong, Rui Yang, and Ruoyu Yang from State
Key Laboratory for Novel Software Technology, Nanjing University, China for the
paper “Continuous Motion Recognition in Depth Camera Based on Recurrent Neural
Networks and Grid-based Average Depth.”
The best presentation paper award went to Ryo Furukawa, Masahito Naito, Daisuke
Miyazaki, Masahi Baba, Shinsaku Hiura, Yoji Sanomura, Shinji Tanaka, and Hiroshi
Kawasaki from Hiroshima City University, Japan, Hiroshima University Hospital,
Japan, and Kyushu University, Japan, for the paper “Auto-Calibration Method for
Active 3D Endoscope System Using Silhouette of Pattern Projector.”
The best paper runner-up award went to Guangbin Wu, Weishan Chen, Wangmeng
Zuo, and David Zhang from the State Key Laboratory of Robotics and System, Harbin
VI Preface

Institute of Technology, Harbin, China, for the paper “Unsupervised Domain Adap-
tation with Robust Deep Logistic Regression.”
The social program of PSIVT 2017 included a banquet at the Xiongchu Interna-
tional Hotel, and a bus excursion to selected sites in Wuhan, followed by a stunning
“survivors’ party” in a nicely illuminated restaurant on the waters of the East Lake.
These social events certainly also contributed to the success of the conference.
We thank our PSIVT 2017 sponsors, the National Engineering Research Center for
E-learning, a joint institute of Central China Normal University and Wollongong
University, and Wuhan Jingtian Electrical Co., Ltd.
We acknowledge the careful work of all reviewers and area chairs, the local
Organizing Committee, especially the general chair, Xinguo Yu, and the head of the
local committee, Chao Sun, both from Central China Normal University; we also thank
this university for providing the conference venue. Of course, last but not least, we
thank the PSIVT Steering Committee that was chaired during 2015–2017 by Reinhard
Klette, Auckland University of Technology, New Zealand, for helpful support
throughout the preparation of PSIVT 2017.
We thank the IAPR, the International Association for Pattern Recognition, for
endorsing PSIVT 2017, and Springer’s Lecture Notes in Computer Science team,
especially Alfred Hofmann and Anna Kramer, for the efficient communication when
submitting and finalizing this volume.
We look forward to seeing you all again at PSIVT 2019 in Sydney, Australia.

December 2017 Manoranjan Paul

Carlos Hitoshi
Qingming Huang
 Organization

Hosted by

Central China Normal University, Wuhan, China

Honorary Chairs
Zhengyou Zhang Microsoft, Redmond, USA
Reinhard Klette Auckland University of Technology, New Zealand

General Chairs
Larry Davis University of Maryland, USA
Antonio NICTA, Australia
Robles-Kelly
Xinguo Yu Central China Normal University, China

Program Chairs
Qingming Huang University of Chinese Academy of Sciences, China
Carlos Hitoshi University of São Paulo, Brazil
Simon Lucey CSIRO ICT Centre, Australia
Manoranjan Paul Charles Sturt University, Australia

General Workshop Chairs

Shin’ichi Satoh National Institute of Informatics, Japan
Chong-Wah Ngo City University of Hong Kong, SAR China
Junsong Yuan Nanyang Technological University, Singapore

Local Arrangements Chairs

Jingying Chen Central China Normal University, China
Chao Sun Central China Normal University, China

Demo/Exhibition Chairs
Tiziana D’Orazio Institute for Signal and Image Processing, Italy
Wen-Huang Cheng Research Center for Information Technology Innovation,
Academia Sinica, Taiwan
Lingyu Duan Peking University, China
VIII Organization

Publicity Chairs
Yong Man Ro Korea Advanced Institute of Science and Technology,
South Korea
Jun-Wei Hsieh National Taiwan Ocean University, Taiwan
Liyuan Li Institute for Infocomm Research, Singapore
Hanqing Lu National Laboratory of Pattern Recognition, CAS, China

Sponsorship Chairs
Bin He Central China Normal University, China
Yue Wang Institute of Information Technology, Singapore

Steering Committee
Chair
Reinhard Klette Auckland University of Technology, New Zealand

Members
Akihiro Sugimoto National Institute of Informatics, Japan
Xinguo YU Central China Normal University, China
Kap Luk Chan Nanyang Technological University, Singapore
Domingo Mery Pontificia Universidad Católica, Chile
Yo-Sung Ho Gwangju Institute of Science and Technology, South Korea
Wen-Nung Lie National Chung Cheng University, Taiwan
Mariano Rivera Centro de Investigación en Matemáticas, México

Area Chairs
Fay Huang National Ilan University, Taiwan
Michael Cree The University of Waikato, New Zealand
Nicolai Petkov Groningen University, The Netherlands
Phil Bones University of Canterbury, New Zealand
Li Chen University of the DC, USA
Jian Cheng Chinese Academy of Sciences, China
Uwe Franke Daimler A. G., Germany
Hanseok Ko Korea University, South Korea
Chilwoo Lee Chonnam National University, South Korea
Wen-Nung Lie National Chung Cheng University, Taiwan
Chia-Yen Chen National University of Kaohsiung, Taiwan
Rick Millane University of Canterbury, New Zealand
Takeshi Oishi Tokyo University, Japan
Lei Qin Chinese Academy of Sciences, China
Terence Sim National University of Singapore, Singapore
Zhixun Su Dalian University of Technology, China
Wei Qi Yan Auckland University of Technology, New Zealand
 Organization IX

Program Committee
Ajjen Joshi Boston University, USA
Alex Carneiro University of São Paulo, Brazil
Alexandre University of Sao Paulo, Brazil
Morimitsu
Andrew Kurauchi University of São Paulo, Brazil
Antonio Diaz Tula University of São Paulo, Brazil
Carlos Elmadjian University of São Paulo, Brazil
Carlos Eduardo Centro Universitario da FEI, Brazil
Thomaz
Chang-Tsun Li Charles Sturt University, Australia
Chia-Hung Yeh National Sun Yat-sen University, Taiwan
Chih-Yang Lin Yuan Ze University, Taiwan
Ching-Chun Huang National Chung Cheng University, Taiwan
Chiung-Yao Fang National Taiwan Normal University, Taiwan
D. M. Motiur Charles Sturt University, Australia
Rahaman
Darin O’Keeffe Christchurch Hospital, New Zealand
Darko Brodic University of Belgrade, Serbia
Deng-Yuan Huang Dayeh University, Taiwan
Elham Saraee Boston University, USA
Estephan Dazzi Federal University of Espirito Santo, Brazil
Wandekoken
Fabricio Martins Federal University of Technology – Paraná, Brazil
Lopes
Fay Huang Ilan University, Taiwan
Filipe Costa Federal University of Minas Gerais, Brazil
Flavio Coutinho University of Sao Paulo, Brazil
Francisco Rodrigues University of São Paulo, Brazil
Frank Borsato Federal University of Technology – Parana, Brazil
Guo-Shiang Lin Dayeh University, Taiwan
Guorong Li Chinese Academy of Sciences, China
Hajime Nagahara Osaka University, Japan
Harvey Ho University of Auckland, New Zealand
Hélio Pedrini University of Campinas, Brazil
Henrique Morimitsu Inria – Grenoble, Switzerland
Hiroshi Kawasaki Kyushu University, Japan
Igor Leonardo Universidade Federal da Bahia, Brazil
Oliveira Bastos
Jasmine Seng Charles Sturt University, Australia
Jesús Mena-Chalco Federal University of ABC, Brazil
Jian Sun Xi’an Jiaotong University, China
Jiunn-Lin Wu National Chung Hsing University, Taiwan
Junbiao Pang Beijing University of Technology, China
Kar-Ann Toh Yonsei University, Japan
X Organization

Keita Takahashi Nagoya University, Japan

Li-Minn Ang Charles Sturt University, Australia
Laszlo Szilagyi Sapientia, Hungarian Science University of Transylvania,
Hungary
Lei Qin Institute of Computing Technology Chinese Academy
of Sciences, China
Li Chen University of the District of Columbia, USA
Li-Wei Kang National Yunlin University of Science and Technology, China
Liang Li University of Chinese Academy of Sciences, China
Lihong Zheng Charles Sturt University, Australia
Marco Porta University of Pavia, Italy
Margrit Betke Boston University, USA
Mohammad Zavid ISI Foundation, Italy
Parvez
Nina Hirata University of São Paulo, Brazil
Niras C. Vayalil Macquarie University, Australia
Norihiko Kawai Nara Institute of Science and Technology, Japan
Pallab Podder Charles Sturt University, Australia
Pascal Monasse Université Paris-Est, France
Qianqian Xu Chinese Academy of Sciences, China
Rafael Will Macedo University of São Paulo, Brazil
de Araujo
Raphael Prates Federal University of Minas Gerais, Brazil
Roberto Cesar University of São Paulo, Brazil
Ryoma Bise Kyushu University, Japan
Sabih Rehman Charles Sturt University, Australia
Sergio A. Velastin Universidad Carlos III de Madrid, Spain
Shampa Shahriyar Monash University, Australia
Sheng-Hua Zhong Shen Zhen University, China
Shih-Wei Sun Taipei National University of the Arts, Taiwan
Shiqian Wu Wuhan University of Science and Technology, China
Shohei Nobuhara Kyoto University, Japan
Shyi-Chyi Cheng National Taiwan Ocean University, Taiwan
Simina Emerich Technical University of Cluj-Napoca, Romania
Takahiro Okabe Kyushu Institute of Technology, Japan
Tariq Khan The University of Auckland, New Zealand
Ting-Lan Lin Chung Yuan Christian University, Taiwan
Tingting Jiang Peking University, China
Tomokazu Sato NAIST, Japan
Toufique Soomro Charles Sturt University, Australia
Weiqi Yan Auckland University of Technology, New Zealand
Wangmeng Zuo Harbin Institute of Technology, China
Wei-Ta Chu National Chung Cheng University, Taiwan
Wen-Nung Lie National Chung Cheng University, Taiwan
Xiaoming Zheng Charles Sturt University, Australia
Xiaoyi Jiang University of Münster, Germany
 Organization XI

Xiujuan Chai ICT, Charles Sturt University, Australia

Xufeng Lin Charles Sturt University, Australia
Yasushi Makihara Osaka University, Japan
Yongkang Wong National University of Singapore
Yongxi Lu University of California San Diego, USA
Zhixun Su Dalian University of Technology, China

Sponsors

National Engineering Research Centre for E-Learning

Central China Normal University Wollongong Joint Institute
IAPR
Wuhan Jingtian Electrical Co. Ltd.

Additional Reviewers

Hossain, Mohammad
Kashiwabara, Andre
Rahaman, D. M. Motiur
Rahman, Hamidur
Ren, Dongwei
Vicente, Fábio
Wang, Hao
Yang, Dong
Yu, Xiangyu
Zhang, Kai
 Contents

Image/Video Processing and Analysis

Multiset Canonical Correlation Analysis: Texture Feature Level Fusion

of Multiple Descriptors for Intra-modal Palmprint Biometric Recognition . . . . 3
Raouia Mokni, Anis Mezghani, Hassen Drira, and Monji Kherallah

Complex-Valued Representation for RGB-D Object Recognition. . . . . . . . . . 17

Rim Trabelsi, Issam Jabri, Farid Melgani, Fethi Smach,
Nicola Conci, and Ammar Bouallegue

Video Highlight Detection via Deep Ranking Modeling . . . . . . . . . . . . . . . . 28

Yifan Jiao, Xiaoshan Yang, Tianzhu Zhang, Shucheng Huang,
and Changsheng Xu

Ink-Jet Printer’s Characterization by 3D Gradation Trajectories

on an Equidistant Color Difference Basis . . . . . . . . . . . . . . . . . . . . . . . . . . 40
Oleg Milder and Dmitry Tarasov

Gradation Surfaces as a Method for Multi-color Ink-Jet Printers

Color Specifications Management . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 53
Oleg Milder and Dmitry Tarasov

Multi-objective Visual Odometry . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 62

Hsiang-Jen Chien, Jr-Jiun Lin, Tang-Kai Yin,
and Reinhard Klette

Visual Comparison Based on Multi-class Classification Model . . . . . . . . . . . 75

Hanqin Shi and Liang Tao

System Designs for Augmented Reality Based Ablation

Probe Tracking . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 87
Hao Bo Yu and Harvey Ho

Using Sparse-Point Disparity Estimation and Spatial Propagation

to Construct Dense Disparity Map for Stereo Endoscopic Images . . . . . . . . . 100
Wen-Nung Lie, Hsi-Hung Huang, Shih-Wei Huang,
and Kai-Che Liu

Globally Optimal Object Tracking with Complementary Use of Single

Shot Multibox Detector and Fully Convolutional Network . . . . . . . . . . . . . . 110
Jinho Lee, Brian Kenji Iwana, Shouta Ide, Hideaki Hayashi,
and Seiichi Uchida
XIV Contents

Single Image Dehazing via Image Generating . . . . . . . . . . . . . . . . . . . . . . . 123

Shengdong Zhang, Jian Yao, and Edel B. Garcia

Automatic Brain Tumor Segmentation in Multispectral MRI Volumes

Using a Random Forest Approach. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 137
Zoltán Kapás, László Lefkovits, David Iclănzan, Ágnes Győrfi,
Barna László Iantovics, Szidónia Lefkovits, Sándor Miklós Szilágyi,
and László Szilágyi

A New Scheme for QoE Management of Live Video Streaming

in Cloud Environment . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 150
Dheyaa Jasim Kadhim, Xinguo Yu, Saba Qasim Jabbar, Yu Li,
and Wenxing Luo

Hybrid Adaptive Prediction Mechanisms with Multilayer Propagation

Neural Network for Hyperspectral Image Compression . . . . . . . . . . . . . . . . 162
Rui Xiao and Manoranjan Paul

Secret Image Sharing for (k, k) Threshold Based on Chinese

Remainder Theorem and Image Characteristics . . . . . . . . . . . . . . . . . . . . . . 174
Xuehu Yan, Yuliang Lu, Lintao Liu, Song Wan, Wanmeng Ding,
and Hanlin Liu

Detection of Adulteration in Red Meat Species Using Hyperspectral

Imaging . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 182
Mahmoud Al-Sarayreh, Marlon M. Reis, Wei Qi Yan,
and Reinhard Klette

Pattern Recognition and Applications

Unsupervised Domain Adaptation with Robust Deep

Logistic Regression . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 199
Guangbin Wu, Weishan Chen, Wangmeng Zuo,
and David Zhang

Continuous Motion Recognition in Depth Camera Based

on Recurrent Neural Networks and Grid-based Average Depth . . . . . . . . . . . 212
Tao Rong, Rui Yang, and Ruoyu Yang

Auto-calibration Method for Active 3D Endoscope System Using

Silhouette of Pattern Projector . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 222
Ryo Furukawa, Masahito Naito, Daisuke Miyazaki,
Masahi Baba, Shinsaku Hiura, Yoji Sanomura, Shinji Tanaka,
and Hiroshi Kawasaki
 Contents XV

A Novel No-reference Subjective Quality Metric for Free Viewpoint

Video Using Human Eye Movement . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 237
Pallab Kanti Podder, Manoranjan Paul, and Manzur Murshed

Research on Color Space Conversion Model from CMYK

to CIE-LAB Based on GRNN . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 252
Xinyue Bao, Wangan Song, and Sheng Liu

GCP-SLAM: LSD-SLAM with Learning-Based Confidence Estimation . . . . . 262

Aidi Feng, Weiqi Zhang, Zifei Yan, and Wangmeng Zuo

Layer-Wise Weight Decay for Deep Neural Networks . . . . . . . . . . . . . . . . . 276

Masato Ishii and Atsushi Sato

Adaptive Dehaze Method for Aerial Image Processing. . . . . . . . . . . . . . . . . 290

Rong-qin Xu, Sheng-hua Zhong, Gaoyang Tang, Jiaxin Wu,
and Yingying Zhu

Efficient GPU Implementation of Informed-Filters for Fast Computation . . . . 302

Takuro Oki and Ryusuke Miyamoto

Automatic Problem Understanding from Circuit Schematics . . . . . . . . . . . . . 314

Xinguo Yu, Pengpeng Jian, Bin He, Gang Zhao, and Meng Xia

Feature Similarity and Frequency-Based Weighted Visual Words

Codebook Learning Scheme for Human Action Recognition. . . . . . . . . . . . . 326
Saima Nazir, Muhammad Haroon Yousaf, and Sergio A. Velastin

A Critical Review of the Trifocal Tensor Estimation . . . . . . . . . . . . . . . . . . 337

Laura F. Julià and Pascal Monasse

Deep Learning-Based Improved Object Recognition in Warehouses. . . . . . . . 350

Syeda Fouzia, Mark Bell, and Reinhard Klette

Understanding Plane Geometry Problems by Integrating Relations

Extracted from Text and Diagram . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 366
Wenbin Gan, Xinguo Yu, Chao Sun, Bin He, and Mingshu Wang

Lane Detection Based on Road Module and Extended Kalman Filter. . . . . . . 382
Jinsheng Xiao, Li Luo, Yuan Yao, Wentao Zou, and Reinhard Klette

Gaussian Noise Detection and Adaptive Non-local Means Filter . . . . . . . . . . 396

Peng Chen, Shiqian Wu, Hongping Fang, Bin Chen,
and Wei Wang

Uncertainty Model for Template Feature Matching . . . . . . . . . . . . . . . . . . . 406

Hongmou Zhang, Denis Grießbach, Jürgen Wohlfeil,
and Anko Börner
XVI Contents

Age Estimation with Local Ternary Directional Patterns. . . . . . . . . . . . . . . . 421

Raphael Angulu, Jules R. Tapamo, and Aderemi O. Adewumi

Computer Vision and Applications

Predicting Turn-Taking by Compact Gazing Transition Patterns

in Multiparty Conversation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 437
Li Tian, Qi Jia, and Zhen Zhu

Intelligent Assistant for People with Low Vision Abilities . . . . . . . . . . . . . . 448

Oleksandr Bogdan, Oleg Yurchenko, Oleksandr Bailo,
Francois Rameau, Donggeun Yoo, and In So Kweon

Using Facial Expression Recognition for Crowd Monitoring. . . . . . . . . . . . . 463

Ross Philip Holder and Jules-Raymond Tapamo

Block-Wise Gaze Estimation Based on Binocular Images. . . . . . . . . . . . . . . 477

Xuemei Wu, Jing Li, Qiang Wu, Jiande Sun, and Hua Yan

Integrated Multi-scale Event Verification in an Augmented

Foreground Motion Space . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 488
Qin Gu, Jianyu Yang, Wei Qi Yan, and Reinhard Klette

Author Index . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 501

Image/Video Processing and Analysis
 Multiset Canonical Correlation Analysis:
Texture Feature Level Fusion of Multiple
Descriptors for Intra-modal Palmprint
Biometric Recognition

Raouia Mokni1(B) , Anis Mezghani2 , Hassen Drira3 , and Monji Kherallah4

1
Faculty of Economics and Management of Sfax, University of Sfax, Sfax, Tunisia
raouia.mokni@gmail.com
2
University of Sfax, Sfax, Tunisia
anis.mezghani@gmail.com
3
Institut Mines-Télécom/Télécom Lille, CRIStAL (UMR CNRS 9189),
Lille, France
hassen.drira@telecom-lille.fr
4
Faculty of Sciences of Sfax, University of Sfax,
Road Soukra Km 3, 3038 Sfax, Tunisia
monji.kherallah@fss.usf.tn

Abstract. This paper describes a novel intra-modal feature fusion for

palmprint recognition based on fusing multiple descriptors to analyze the
complex texture pattern. The main contribution lies in the combination
of several texture features extracted by the Multi-descriptors, namely:
Gabor Filters, Fractal Dimension and Gray Level Concurrence Matrix.
This means to their effectiveness to confront the various challenges in
terms of scales, position, direction and texture deformation of palmprint
in unconstrained environments. The extracted Gabor filter-based texture
features from the preprocessed palmprint images to be fused with the
Fractal dimension-based-texture features and Gray Level Concurrence
Matrix-based texture features using the Multiset Canonical Correlation
Analysis method (MCCA). Realized experiments on three benchmark
datasets prove that the proposed method surpasses other well-known
state of the art methods and produces encouraging recognition rates
by reaching 97.45% and 96.93% for the PolyU and IIT-Delhi Palmprint
datasets.

Keywords: Palmprint · Texture analysis · Gabor Filters

Fractal Dimension · Gray Level Concurrence Matrix
Information fusion · Multiset Canonical Correlation Analysis (MCCA)

1 Introduction and related works

Human biometrics has become an area of tremendous importance and poten-
tial. By human biometrics we mean the use of physiological characteristics,
c Springer International Publishing AG, part of Springer Nature 2018
M. Paul et al. (Eds.): PSIVT 2017, LNCS 10749, pp. 3–16, 2018.
https://doi.org/10.1007/978-3-319-75786-5_1
4 R. Mokni et al.

of human body parts and their appearances, to identify individual human beings
in the course of their daily activities. The appearances of body parts, especially in
imaged data, have a large variability and are influenced by their texture patterns,
colors, lighting environment, and so on. Therefore, the biometrics researchers
have focused on body parts and images that try to minimize this variability
within class (subjects) and maximize it across classes. This paper investigates
the identification of the persons from one of the physical body parts, namely:
Palmprint. Till now, more attention by many researchers has been paid to palm-
print recognition systems thanks to its low cost, affordable materials, high recog-
nition accuracy, etc. These systems have been grouped into two major ways: (1)
Line-based structural approaches and (2) Texture-based global approaches.
Line-based structural approaches group essentially the principal lines [20], the
wrinkles [4] and the ridges [5]. However, the major drawback of these approaches
is the presence of the computation cost to extract its features, which means to
a significant computational time. Unluckily, those features alone cannot provide
satisfactory information to identify the person efficiently. Otherwise, texture-
based global approaches incorporate the global information of images involving
the texture. Various descriptors have been applied for analyzing the palmprint
texture pattern, such as the Local Binary Pattern (LBP) [11], Fractal Dimension
(FD) [7], Discrete cosine Transform (DCT) [2], Gabor Wavelets [28], Gabor filter
(GF) [16], Gray Level Co-occurrence Matrix (GLCM) [19], etc.
Many researchers claim that they spend less effort on analyzing the texture
information contrast the methods that are relied on analyzing the principal line
shape information. Therefore, the accuracy of their recognition commonly relies
on the reliability of the features.
In this paper, we propose a novel intra-modal feature fusion approach for
palmprint identification based on fusing multiple-texture features extracted by
different descriptors using the feature fusion level. For this end, Gabor filter,
Fractal Dimension and Gray Level Co-occurrence Matrix descriptors are used to
analyze the palmprint texture. Subsequently, we apply the dimensionality reduc-
tion technique based on the Principal Component Analysis in order to select the
more discriminant feature. Then, we fuse these obtained features at feature fusion
level relying on the Multiset Canonical Correlation Analysis method.
The reminder of this paper is structured as follows: In Sect. 2, the pipeline of
the proposed intra-modal feature fusion approach and its modules are detailed.
Experimental evaluation and comparison of the accuracy of our proposed app-
roach with other state-of-the-art methods are presented in Sect. 3. Lastly, a con-
clusion and some perspectives are drawn in Sect. 4

2 Proposed Intra-modal Texture-Feature Fusion

for Palmprint Identification
The pipeline of our suggested intra-modal feature fusion approach for palm-
print identification summarizes the following modules: (1) Hand Acquisition
 Multiset Canonical Correlation Analysis: Feature Level Fusion 5

(2) Palmprint Preprocessing (3) Multiple-Texture-Feature Extraction (4) Fea-

ture Selection and Fusion (5) Matching and (6) Decision.
The whole modules of our proposed approach is depicted in Fig. 1.

Fig. 1. Flowchart of our proposed intra-modal palmprint identification approach based

on multiple-texture features

2.1 Palmprint Preprocessing

The palmprint preprocessing is a paramount module in the general biometric

system. It is devoted for extracting the Region of interest (ROI) which has dis-
criminatory and important features of recognition. This module includes three
major steps, namely: Edge detection of the hand, Coordinate system Stabiliza-
tion and ROI Preprocessing, which are described in detail in our previous work
[23]. These steps are summarized in the following items:

– (1) Edge detection of the hand: This step aims to detect the hand contour
in order that we can detect the finger-webs addressing to stabilize the ROI.
To this end, we firstly converted the input hand image into binary image
using an automatic thresholding technique such as the Otsu technique [24].
Then, we applied the Morphology operations incorporating the closing and
the opening operations in order to surmount some identification problems,
which are founded when the users place their hands in front of the camera
with wearing accessories as the rings. Finally, we detected the contour of the
hand and applied the smoothing filter.
– (2) Coordinate System Stabilization: In order to extract the ROI, it is
necessary to stabilize the coordinate system. For this end, we start by adjust-
ing the hand image with its edge and detecting its centroid as a reference
point. Subsequently, we extract the finger-webs in order to obtain a stable
coordinate system which enables us to extract and locate the ROI square.
6 R. Mokni et al.

– (3) ROI Preprocessing: After the ROI square extraction and localization,
it is necessary to fix its size to S × S with S= 128 pixels and then rotate it to
a vertical position. Lastly, we applied a low pass filter to enhance the quality
of ROI in order to restrict the noise.

Figure 2 shows the original image, the edge detection of the hand, the stabiliza-
tion of the coordinate system and the final extracted ROI of the palmprint.

Fig. 2. The Palmprint Preprocessing: (1) Edge detection of the hand, (2) Coordinate
System Stabilization and (3) ROI Preprocessing

2.2 Multiple-Texture Feature Extraction

In this section, we provide a set of different features for every ROI palmprint
image using multiple widely descriptors that better characterize the texture pat-
tern in unconstrained environment, such as the Gabor Filter (GF) [18], the
Fractal dimension (FD) [6] and the Gray Level Concurrence Matrix (GLCM)
[12] descriptors. These different descriptors show its robustness to address the
several challenges in terms of scales, position, direction and texture deformation
of palmprint, that are produced when acquiring the images.

The Gabor Filter Descriptor. The Gabor Filter (GF) is one of the earliest
and valuable frequency methods used to analyze the texture of the image. It
has a huge variety in various pattern recognition applications [9,10,19]. The
mainly advantage of GF focuses on their different challenging and variability
concerning the rotation, position, translation, scale and lighting conditions of
palmprint texture image in unconstrained environments. The Gabor filter is
based on extracting directly the texture features from the Gray level images.
In the spatial domain, a 2-D Gabor filter based in both frequency (scale) and
orientation is a Gaussian kernel function modulated by a complex sinusoidal
plane wave, defined as the following equation:
 Multiset Canonical Correlation Analysis: Feature Level Fusion 7
   

μ2 x 2 + γ2 × y 2 
G(x, y) = exp − exp(j × 2 × Π × μ × x + φ);
Π ×γ×η 2σ 2

x = x cos ν + y sin ν

y = −(x sin ν + y cos ν)
(1)
Where μ represents the central frequencies (scale) of the sinusoidal plane
wave and ν is the orientation of the normal to the parallel stripes of a Gabor
function. σ is the standard deviation of the Gaussian envelope, φ is the phase
offset and γ is the ratio of the spatial aspect that stipulates the ellipticity of the
Gabor function support.
In our experimental evaluations, the palmprint image is convolved with
G = 40 Gabor filters at five frequencies (μ = 5) and eight orientations (ν = 8),
as illustrated in Fig. 3. We pursue to apply the same Gabor filters parameters
suggested by Haghighat et al. [10] to build the feature vector. Thus, let us define
that we calculated the vector dimension as:

F VGF = ((S × S) × μ × ν)/(d × d) (2)

Where S = 128 pixels is the size of palmprint ROI image, μ = 5 is the

number of frequencies, ν = 8 is the number of orientations and d = 4 is the
factor of downsampling the features that is used to reduce the redundancy of
the texture information producing by the correlation between the adjacent pixels
in palmprint ROI image [10]. Therefore, the dimension of the feature vector is
equal to (128 × 128 × 40)/(4 × 4) = 655,360/16 = 40,960. In addition, this vector
is normalized to zero mean and unit variance.

Fractal Dimension Descriptor. Fractal Dimension (FD) method is one of the

effective methods used to extract the Model based texture features. It is consid-
ered as a widely applied descriptor, especially for analyzing the texture represen-
tation, in several fields like the signature recognition [30], palmprint recognition
[7,22], writer identification [3], etc. A fractal object has a very complicate shape

Fig. 3. (a) The extracted palmprint ROI. (b) The real part with applying the Gabor
filter in five frequencies and eight orientations on the palmprint image.
8 R. Mokni et al.

which is a collection of different sub-objects recurring in different scales. Several

methods are proposed to compute the fractal dimension like the Box Counting
(BC), Differential Box Counting (DBC), Mass Radius (MR), Cumulative inter-
section (CI), etc. Among these methods, the BC is the deeply used method to
estimate the fractal dimension and analyze the texture shape thanks to its utility
to describe the interruption and complexity of the image texture. The ultimate
principle of BC method focuses on converting the input image to binary image
and then splitting it into a boxes number with several scales . After that, it is
necessary to compute the number of boxes including the shape information N ()
for each scale. Thus, the fractal dimension F D is obtained by a least squares
regression method, as illustrated in the following equation:
log(N ())
F D = slop (3)
log(1/)
In our experiments, we proposed a method to provide the shape information
in horizontal and vertical spaces of palmprint ROI image. In this case, we divided
it, sequentially, into r × r sub regions equivalent squares. It is imminent that the
value r =10 is chosen after a set of experiments. For each sub-region, the fractal
Dimension F D is calculated. Therefore, we obtained a sequence of F Ds for the
whole input palmprint ROI image presented by the following feature vector:

F VF D = F D1 , F D2 , . . . , F D(r×r) (4)

Gray Level Concurrence Matrix Descriptor. The Co-occurrence Matrix

(GLCM) is one of the robust statistical methods used to describe the image
texture [19]. This method presents a matrix which contains valuable information
about the statistical distribution of intensities and provides information about
the relative position of neighborhood pixels of the analyzed image.
More specifically, let us have an image Img, of size S × S, the GLCM matrix
Gm can be defined by Riabaric et al. [25] as: Gm(i, j, δ, θ), size G × G, where the
values of the pixels are quantized into G levels, contains at the position (i, j) a
number of occurrences of a pair of pixels that are at the offset δ = 1, 2, 3, . . . , etc,
(the offset δ is a distance between the interested neighborhood pixels) in the
direction θ = 0◦ , 45◦ , 90◦ , 135◦ (θ is the angle between interested neighborhood
pixels or the rotation angle of an offset), where one pixel has a gray-level value
i and another pixel has a gray-level value j. In literature, Haralick described
fourteen statistical Haralick features to normalize the gray level co-occurrence
matrices [12]. In our empirical experiments, we applied only six of these features
which are:
– Contrast:
G−1
 G−1
 G−1

f1 = n2 Gm(i, j);
n=0 i=0 j=0 (5)
n = |i − j|
 Multiset Canonical Correlation Analysis: Feature Level Fusion 9

– Variance:
G−1
 G−1

f2 = (i − μ)2 Gm(i, j) (6)
i=0 j=0

– Energy:
G−1
 G−1

f3 = Gm(i, j)2 (7)
i=0 j=0

– Entropy:
G−1
 G−1

f4 = Gm(i, j) × log(Gm(i, j)) (8)
i=0 j=0

– Homogeneity or Inverse difference moment:

G−1
 G−1
 Gm(i, j)
f5 = (9)
i=0 j=0
1 + (i − j)2

Where Gm is GLCM matrix and G is Grey Scale value, μ is the mean value.
– Correlation:
G−1
 G−1
 (i × j) × Gm(i, j) − (μx × μy )
f6 = (10)
i=0 j=0
σx σy

Where Gm is GLCM matrix and G is Grey Scale value, μx , μy are the mean
value and σx , σy are standard deviation along two axes X and Y .
In this work, we follow the same parameters proposed in our prior work
[19] to construct the feature vector. So, the number of gray levels of palmprint
ROI is G = 256. We have split each ROI into small 8 × 8 sub-images by a
non-overlapping sliding window. For each ROI, we have obtained 64 sub-images.
For each sub-image, the GLCMs matrices are calculated corresponding to four
different values of direction with four different offset values mentioned above.
Hence, we have obtained 16 GLCMs (4 (offset value) × 4 (direction number))
for each sub-image. Then, we have calculated the six Haralick features for each
GLCM. Therefore, the feature vector is calculated by the multiplication of the
number of GLCMs for ROI image and the number of Haralick features, as follows:

V FGLCM = (64 × 16) × 6 = 6144. (11)

2.3 Feature Selection and Fusion

The fusion of different information in the biometric system has gained a great
interest in various researchers teams. This fusion between the multiple descrip-
tors will be improve the recognition accuracy. Indeed, it can occur in different
10 R. Mokni et al.

ways for a recognition system, such as: Fusion at the data or feature level, Fusion
at the score level and Fusion at the decision level.
Several prior researchers believe that the fusion at feature level way is con-
sidered to be more effective than the other ways by virtue of the capability of
fusing the multiple sources of information to achieve a single feature set, that
contains more efficient richer information than the fusion at the score and the
decision levels [8,17]. Thus, the fusion at feature level is widely important to
produce a substantial identification. There are many manners of the fusion at
the feature level [29], namely: the Serial feature fusion and the Parallel feature
fusion. The two fusion manners can ameliorate the accuracy of recognition of
persons to some extent. However, neither of these fusion methods focused on the
inherent correlations across multiple feature sets. Therefore, both of these fusion
methods cannot reach tremendous feature fusion. To surmount this weakness,
the Canonical Correlation Analysis (CCA) feature fusion method is conducted
to extract the feature pairs describing the intrinsic relationship between two fea-
ture sets [21,26] and also the Multiset Canonical Correlation Analysis (MCCA)
method is considered as a generalization of the CCA method and is proposed to
be suitable for fusing more than two sets of features, as presented in our cases.

Fig. 4. Multiset Canonical Correlation Analysis method for 4 feature vectors with
rank(F V1 ) > rank(F V2 ) > rank(F V3 ) > rank(F V4 )

Therefore, the main principle of MCCA method is conducted by combining

the first two feature vectors that have highest ranks using CCA method and
then fusing the result with the next feature vector that has an utmost rank
using CCA method, and so on. Figure 4 illustrates an example of the process
of the MCCA method for four feature vectors with rank(F V1 ) > rank(F V2 ) >
rank(F V3 ) > rank(F V4 ). In this example, the first step of MCCA focuses on
fusing F V1 and F V2 , which have the highest ranks using CCA method. Then,
F V12 is fused with the next highest rank feature vector F V3 and finally, F V4 is
fused with F V123 using yet CCA method.
 Multiset Canonical Correlation Analysis: Feature Level Fusion 11

A Brief Background of Canonical Correlation Analysis Method. The

Canonical Correlation Analysis method firstly implements a function of corre-
lation criterion between two sets of feature vectors and secondly calculates the
projection vector set of these two sets depending on the criterion. Finally, the
fused canonical correlation features are extracted.
As defined in [21], given X ∈ Ra×n and Y ∈ Rb×n two matrices, where n is
the number of samples, a is the dimensionality of the first feature set and b is
the dimensionality of the second feature set.
Suppose that Mxx ∈ Ra×a and Myy ∈ Rb×b present the within-sets covariance
matrices of X and Y . Additionally, Mxy ∈ Ra×b the matrix defines the between-
T
set covariance matrix (where Myx = Mxy ). Hence, M includes all the information
about the relationship between pairs of features, as defined follows:
   
Mxx Mxy cov(x) cov(x, y)
M= = (12)
Myx Myy cov(y, x) cov(y)

Although the correlation between the two feature vector sets shows no coher-
ent pattern, there exist huge difficulties to understand and analyze the associa-
tion between multiple sets from this matrix [8,15,21]. For this purpose, the CCA
method aims to find the linear combinations, X ∗ = WxT X and Y ∗ = WyT Y ,
which maximize the pair-wise correlation through the two data sets. Wx and Wy
are the transformation matrices which are establish by obtaining the following
eigenvalue equations:
 −1 −1
Mxx Mxy Myy Myx Ŵx = ξ 2 Ŵx
−1 −1
(13)
Myy Myx Mxx Mxy Ŵy = ξ 2 Ŵy

Where, Ŵx and Ŵy are the eigenvectors and ξ 2 is the diagonal matrix of
eigenvalues or squares of the canonical correlations. These two matrices Wx
and Wy consist of the classed eigenvectors corresponding to the non-zero eigen-
values. X ∗ , Y ∗ ∈ Rr×n are renown as canonical variates, note that r is the
length of the projected feature vector, which presents the rank of the between-
set covariance matrix Mxy , which is equal to the minimum rank of both input
feature matrices, X and Y . Thus, r = rank(Mxy ) ≤ (min(rank(X), rank(Y ))
i.e. r ≤ rank(n, a, b) is the number of non-zero eigenvalues in each equation,
that will be selected by the decrease order, τ1 ≥ τ2 ≥ τ3 . . . ≥ τr .
In our experiments, in accordance by the suggested work by Sun et al. [26],
the feature fusion is conducted either by summation or concatenation of these
linear transformed feature vectors:
 T  
∗ ∗ T T Wx X
ZSum = X + Y = Wx X + Wy Y = (14)
Wy Y

Or     T  
X∗ WxT X Wx 0 X
ZConcat = = (15)
Y∗ WyT Y 0 Wy Y
12 R. Mokni et al.

where ZSum and ZConcat are named the Canonical Correlation Discriminant
Features by summation and concatenation manners, respectively.
The CCA based fusion has a inconvenient which is focused in a Small Sample
Size (SSS) issue. In the most applications, the number of samples n is habitually
less than the number of features (n < a) and (n < b). In order to overcame
this SSS problem, we proposed to reduce the dimensionality of the obtained fea-
ture vectors (F VGF , F VF D and F VGLCM ) using Principal Component Analysis
method (PCA) [14] and then applied the MCCA method incorporating for each
fusion between two vectors the CCA method using the concatenation manner
defined in Eq. 15 to fuse the different features vectors into a single feature vector.

2.4 Matching
Support Vector Machine (SVM) is a supervised classifier which is a vigorous
statistical technique suggested by Vapnik [27]. This classifier has been widely
intended to resolve the problems of discrimination, classification or regression
in several applications. It is applied to determine an optimal separating hyper-
plane or decision surface with the adaptation of a new corresponding technique
to map the sample points into a high-dimensional feature space and categorized
using a non-linear transformation, even when the data are linearly inseparable.
The optimal hyper-plane is obtained by working out a quadratic programming
problem relied on regularization parameters. This transformation was performed
by kernel functions like Linear Kernel (LK), Radial Basis Function (RBF) and
Polynomial Kernel (PK) which are exploited in this work in order to classify the
query sample.

3 Experiments and Analyses

We performed our experimental researches fusing different texture information
using MCCA method. The proposed approach was assessed over IIT-Delhi [13]
and PolyU [1] databases. The results are discussed below.

3.1 Database Description

The IIT-New Delhi India campus provides a IIT-Delhi palmprint database that
includes 2300 images acquired from 230 persons aged from 12–57 years. For
each person, five samples have been collected from both left and right palms.
These palmprint samples are captured in unconstrained environment with sev-
eral challenges and variation in terms of position, scale, direction and texture
deformation.
The Hong Kong Polytechnic University offered a PolyU-Palmprint database
that contains 7752 images captured from 386 persons. Twenty samples of each
palm were acquired in two separate sessions. Each session contains ten sample
images. The average time interval between these sessions is two month. The
samples collected during these two separate sessions acquired by two different
cameras. This means that there exist a variation in terms of lighting between the
first and second sessions, which leads to more difficulty in person’s identification.
 Multiset Canonical Correlation Analysis: Feature Level Fusion 13

3.2 Experimental Evaluation

We evaluated the accuracy of our suggested approach over two benchmark
databases described above. For the IIT-Delhi database, we arbitrarily singled
three samples of each palm as gallery set and the remainder of samples as the
probe set. On the other hand, we selected ten samples of each person from the
first session of the PolyU database for the gallery set and ten samples from the
second session were chosen for the probe set. This mixture between the samples
collected in separate sessions is important in a real application.
In our experiments, we conducted the performance of our proposed intra-
modal feature fusion using two different methods of fusion at feature levels,
namely: Serial and MCCA methods. For more clearly, the Serial feature fusion is
based on combining multiple feature vector sets into a single feature vector using
a simple concatenation. We recall that before fusing the multiple features by
MCCA method, we applied the dimensionality reduction based on PCA method
in order to overcome the SSS problem. Likewise, when using Serial feature fusion
method, we also used the PCA method for reducing the dimensionality of the
final obtained vector. So, we reveal the experiment results of the Serial and
MCCA feature fusion methods corresponding to several kernel functions of SVM
classifier for IIT-Delhi and PolyU datasets in Table 1. It can be noticed that the
recognition rates obtained by the SVM classifier with Radial Basis Kernel (RBF)
outperforms other kernel functions results (LK and PK) for both datasets.

Table 1. The different results of our proposed intra-modal feature fusion app-
roach based on both Serial and MCCA method evaluating over PolyU and IIT-Delhi
databases.

RRs (%) PolyU-database IIT-Delhi-database

LK RBF PK LK RBF PK
Serial + PCA 96.65 96.95 57.50 94.20 94.50 56.43
PCA + MCCA 97.20 97.45 60.55 96.88 96.93 57.03

Since we performed our experiments with both the MCCA and the Serial
feature fusion methods, it is obvious that the performance of the MCCA feature
fusion method surpasses the Serial feature fusion method for both databases.

4 Comparison with Other Works

Finally, we performed a comparison between our work and some of the earlier
published works in literature, as Highlighted in Table 2. The protocols setting
in terms of the sample classification between both the gallery set and the probe
set used in these different experiments are the same, over IIT-Delhi dataset.
Therefore, it is noted that we achieved a better RR of about 96.93% than both
the presented works in [16,20]. Despite all the experiments over PolyU database
14 R. Mokni et al.

conducted by a mixture of two sessions, which leads to difficulty for individ-

ual’s identification, this protocol is very significant in a real application. Notice-
ably, the result of our proposed method outperforms the totality of prior works
results. For obtaining a fair comparison with the state-of the-art methods, we
yet evaluated our method following the same evaluation protocols presented in
the proposed works by [19,28]. Therefore, our proposed method provided yet
the better performance 97.38% and 97.00% than these works, respectively. This
brings to improved rates of about 0.83 % and 1.60 % to the results of these works
([19,28]).

Table 2. The comparison between our work and some of state of-the-art works. Pro-
tocols description: gallery set (G), probe set (P).

Approaches Palmprint-databases RRs (%)

(protocol setting)
Method in [16] Gabor filters IIT-Delhi (3G/2P) 95.00
Method in [20] FDBC+RF IIT-Delhi (3G/2P) 95.80
Method in [19] GLCM, Gabor Fil- PolyU (10G/5P) 96.55
ters+GDA+SVM
Method in [2] DCT, LDA+WSR PolyU (2G/10P) 97.00
Method in [28] 2DGW+PCNN+ PolyU (5G/10P) 95.40
SVM
The proposed method GF+FD+GLCM+ IIT-Delhi (3G/2P) 96.93
PCA+MCCA+
PolyU (10G/10P) 97.45
SVM (RBF)
PolyU (10G/5P) 97.38
PolyU (5G/10P) 97.00

These achieved performance evaluating over two benchmark datasets demon-

strate the effectiveness of our proposed method to the different existing chal-
lenges in unconstrained environment.

5 Conclusions and Perspectives

In this paper, we presented an intra-modal-texture feature fusion approach based

on fusing multiple feature sets extracted from a texture representation. In fact,
we combined different texture information extracted by multiple descriptors, like
Gabor Filter, Fractal Dimension and Gray Level Concurrence Matrix descriptors
which are invariant to different existing challenges. In fact, we proposed a feature
fusion method such as the Multiset Canonical Correlation Analysis (MCCA) to
fuse these obtained texture information incorporating the Principal Component
Analysis (PCA) in order to reduce the feature dimensionality of each feature set.
This proposed approach prove its robustness against different challenge attacks
Another random document with
no related content on Scribd:
died away. And, greatly to her own astonishment, she found her
head falling heavily upon that of her new friend.
“Ah, child, it is selfish of me to make you talk!” cried Mrs. Dale.
“You are faint, and must rest now. Come and talk to me some other
time.”
Mabin overcame the faintness which had seized her, and quite
suddenly raised her head again. The little excitement of the hope
held out to her brought all her senses back.
“Come and see you! Oh, may I? I should like to so much!”
The girl almost nestled, as she spoke, against her new friend.
But over Mrs. Dale’s fair, childlike face there came at once a sort
of shadow, as if a terrible remembrance had suddenly taken the
power for all pleasurable emotion from her. It almost seemed to
Mabin that the little hands made a movement as if to push her away.
And then there burst forth from the infantile red lips some words
which struck terror into her young hearer, so bitter, so full of sadness,
of biting remorse, were they:
“No, child, no. You must not come. I am too wicked!”
The girl was struck dumb. She wanted to comfort pretty Mrs. Dale;
she wanted to laugh at her self-accusation, to express incredulity,
amusement. But in the face of that look of anguish, of that
inexpressibly mournful cry straight from the heart, she could not
even open her lips. She knew that there was some grief here which
no words of comfort could touch.
So deeply absorbed was she in the silent compassion which kept
her with lowered eyelids and mute lips, that she was quite startled
when Mrs. Dale’s voice, speaking in her ordinary tones, struck again
upon her ear.
“That young fellow who picked you up is one of the Vicar’s sons,
isn’t he?”
“Yes,” answered Mabin in a rather colder voice.
“He seems a very nice lad, and very much interested in—
somebody?” suggested Mrs. Dale archly.
Mabin laughed.
“Yes, so he is. But it is not the ‘somebody’ you mean,” answered
she. “Mrs. Bonnington, that’s his mother, says he can think about
nothing but—Mrs. Dale!”
 Again the sweet face changed; and it was in a low voice full of
sadness that the lady in black said, slowly and deliberately:
“I hope with all my heart that she has made a mistake.” Then, with
a rapid gesture, as if brushing away some thought which was full of
untold terror, she added with a shudder: “Don’t let us talk about it. It
is too horrible!”
 CHAPTER III.
AN INVITATION AND A WARNING.

Mabin’s sprained ankle was a more serious affair than she had
supposed. For a month she never left the house, and for another she
went out in a wheel-chair, or hopped about on a pair of crutches.
And during all that time she caught no glimpse of the pretty
neighbor who had done her such eminent service at the time of the
accident. In vain she had hung about the road outside “The Towers”
looking up at the west side of the house, which was built into the wall
alongside the road, trying to distinguish the fair, blue-eyed face at
one of the windows which peeped sombrely out of the ivy.
Dreary the place looked, Mabin thought, as she pondered over the
mystery surrounding the lady in black. The lowest window visible
from the road was about three feet above the girl’s head; and all she
could see was a pair of crimson moreen curtains, which, she
thought, harmonized ill with what she had seen of the tenant of the
gloomy house. The house had long been “To let, Furnished.” But
why had not dainty Mrs. Dale improved away those curtains?
Mabin did not usually trouble her head about such trifles as
furniture; but she had enshrouded the figure of the pretty widow in
romance; and she felt that her fairy queen was not living up to her
proper standard in contenting herself with crimson moreen.
“What are you looking at so intently?”
Mabin, who, leaning on her crutches, was gazing up at that
mysteriously interesting window, started violently as she saw a white
hand, glistening with diamonds, thrust suddenly out through the ivy
in the midst of a space which she had taken for blank wall.
And, parting the close-growing branches, Mrs. Dale peeped out,
pink and fair and smiling, from a window at the same level as the
one Mabin had been watching, but so thickly covered with ivy that
the girl had not suspected its existence.
“I—I was looking for you. I was hoping to see you,” stammered
Mabin.
 “And now that you have seen me, won’t you please condescend to
see a little more of me?” asked Mrs. Dale. “I won’t eat you up if you
come into my den. Look, here is another inhabitant whom I have
entrapped. But there are strawberries enough for three.”
Mabin hesitated; not from any scruples about the propriety of
visiting the lady about whom so much gossip was talked, certainly,
but because she was shy, and because the thought of a meeting and
a talk with her ideal heroine and a stranger seemed rather
formidable.
But Mrs. Dale would not allow her time to refuse.
“I will send the other inhabitant down to let you in,” said she.
And the ivy closed again, and Mabin could hear the lady’s voice
giving some directions to some person within. She moved
mechanically, on her crutches, toward the high closed gates. And by
the time she reached them they were opening, and Rudolph was
holding them back for her.
The girl could not repress a slight exclamation of astonishment.
Rudolph reddened.
“You are surprised to see me,” said he, rather bashfully. “I hope
you won’t refuse to come in because I am here? I will go away rather
than that.”
Mabin hesitated. She was not very worldly-wise, but it seemed to
her that there was something rather strange about his presence in
the house where the rest of the Vicar’s family were not allowed to
enter. And at the same moment she remembered Mrs. Dale’s
apparent horror at the idea of the young fellow’s admiration for her.
Rudolph’s color deepened still more.
“Why are you always so rude to me, Mabin, or I suppose I ought to
say—Miss Rose?” asked he quickly. “Doesn’t it seem rather unfair,
when you come to think of it? We were great chums once, you
know? Weren’t we?”
“When we were children, yes,” replied Mabin stiffly.
“And why not now?”
The blood rushed to the girl’s forehead.
“How can you ask?” she said, indignantly. “When I owe my
lameness to you?”
Rudolph stared at her, as if uncertain whether he heard aright.
 “To me?”
“Why, yes. Surely you don’t pretend it was not you who threw the
stone which knocked my bicycle over?”
The stiff haughtiness with which she said this melted suddenly into
apologetic alarm when she saw by the change to fierce indignation in
Rudolph that she had made another, and most absurd blunder. At
first he could only stare at her in speechless anger and amazement.
“Do you take me for a street-urchin?” he asked at last.
Mabin recovered herself a little, and refused to be withered up.
“Your brothers do it,” she said below her breath.
“Then I’ll give the little beggars a good hiding the first time I catch
them at it,” said Rudolph sharply. “But I should have thought you
could distinguish the difference between a man and a schoolboy,
and not have visited their sins upon me.”
Mabin felt miserable. She blushed, she stammered when she tried
to speak; and the tears came into her eyes.
“I—I’m sorry!” she said in a constrained voice. “I—I see, I might
have known. But you know—you were rude to me—that very day—
when I saw you at Seagate!”
“Ah! I remember! I asked you to have a cigarette. It was
injudicious, not rude. You should have made a distinction again.”
There was an awkward silence. Rudolph was still resentful; but
when he saw the downcast eyes, and the tears which were
beginning to fringe the long black lashes, he found himself softening.
And, putting her hand too hastily into her pocket for the handkerchief
to wipe away her tears, Mabin dropped one of her crutches.
“Let me help you along,” said he in a gentle voice, as he picked up
the fallen crutch. “I don’t like to see a girl using those things.”
And, without waiting for her permission, he thrust the crutch under
one arm, and insisted on supporting the unwilling girl with the other.
And as they crossed the broad gravelled space to the portico, in the
shade of the trees, Mabin felt a curious sensation of peace and of
pleasure, and suddenly looked up at her companion with a frank
smile.
“I’m very glad we’re friends again,” said she.
And he, smiling too, but with a little more malice, a little more guile,
than she, answered readily:
 “Why, so am I. But I must remind you that it is your fault, not mine,
that we have ever been anything else.”
Mabin hung her head, feeling rather guilty, but with yet more
enjoyment of the present reconciliation than remorse for the past
estrangement. Instead of taking her straight in, Rudolph led her
across the gravel to a flower-border, where, in a little open patch of
sunlight, a rosebush grew. It was a “Mrs. John Lang,” and the huge
pink blossoms were in their full beauty and fragrance.
“I’ve brought you here,” he said didactically, “to read you a moral
lesson. Here we have a rose, full of beauty and sweetness to every
one, but without any thorns. While some Roses I know——”
“Are all thorns to everybody, and are without any beauty,” finished
Mabin for him, laughing, “and without any sweetness.”
“No, no, not at all. But they seldom let you come near enough to
admire the beauty, and they are rather chary of their sweetness.
Now I hope you’ll profit by this lesson.”
“To be sure I—shan’t!” replied Mabin with a rather doleful smile. “I
do try to be less—less objectionable—sometimes,” she added with
seriousness which made Rudolph smile. “But it doesn’t seem to be
very successful. I think I’m going to give up the effort and accept the
fate of ‘an awful example’ as serenely as I can.”
Rudolph tried not to let his smile grow too broad for politeness.
“You are an odd girl,” he said at last. “Or is it another insult to tell
you so?”
Mabin shook her head.
“If it’s an insult,” answered she, “it is one that I’m used to.”
“One is almost as much afraid of saying anything complimentary to
you as of giving you what you call insults,” he began cautiously.
“Otherwise I would tell you that I like ‘odd’ people, people who don’t
always say and do the right thing, that is.”
“Then you ought to appreciate me,” retorted Mabin quickly. “For
everybody says I always do and say the wrong thing!”
“I do appreciate you.”
Mabin laughed and blushed.
“I only said that in fun,” she said awkwardly.
“Well, I said what I did in earnest.”
“Mrs. Dale will be wondering what has become of us,” said Mabin.
 She was not at all anxious to go in; but the pleasure she felt in this
talk with Rudolph had grown rather alarming to her reserve. She
began to fear that she would spoil it all by one of her far-famed
blunders of speech. And so she chose to cut short the enjoyment
while it could remain a recollection of unalloyed delight.
Rudolph, on his side, was in no hurry to go in; although he took a
step obediently toward the portico.
From a feeling of perversity which she could not have accounted
for, Mabin chose to talk about Mrs. Dale as they went slowly toward
the house.
“I have been longing to see her ever since my accident,” she said.
“But although I have been always hovering about the place, wishing
she would come out, to-day is the very first time I have caught sight
of her.”
“That is exactly my own experience,” said Rudolph. “She seems to
have given up driving about the place, and to have shut herself up in
this dreary old house just like a nun.”
“Oh!” said Mabin, feeling quite relieved to hear that he had not, as
she had supposed, been in the society of the beautiful widow
constantly since the day of the accident.
“Yes,” he went on. “I was passing by only half an hour ago, when I
glanced up at the windows, and Mrs. Dale stopped me to ask if I had
heard how you were. And then she asked me in, saying she felt
lonely. And so should I, so would any one, in that mouldy old house
all alone.”
“Poor lady! I am so sorry for her!” said Mabin.
Rudolph looked at her quickly.
“Do you feel like that about her too?” said he earnestly. “All the
other people one meets are either jealous of her beauty, or envious
of her handsome turnout, or angry with her for not wanting to make
their acquaintance.”
“I am very sorry for her,” answered the girl gravely. “I feel certain
that she has had some very great sorrow—”
“Why, yes, her husband’s death,” suggested Rudolph.
“Oh, yes, that of course,” assented Mabin, surprised to find that
the universal doubts whether Mrs. Dale really was a widow had
infected her also. “But something even more than that, I should think.
I have an idea that there is something tragic in her story, if one only
knew it.”
Rudolph said nothing to this, but he looked at his companion with
a quick glance of surprise, as if he himself shared her opinion, and
was astonished to find it echoed.
They were under the portico now, and as their footsteps sounded
on the stone, they saw through the open door into the dark hall and
heard Mrs. Dale’s soft voice calling to them.
“It takes ever so much longer to get a thing done than to do it
one’s self!” she exclaimed brightly, with a sigh, as she came out of
the room on the left, and invited them to go in. “I could have brought
Miss Rose in in half the time, even if she had fought to get away. Did
she fight?” went on Mrs. Dale with arch innocence.
They were in the room by this time, and Mabin, coming in out of
the glare of the sun, stood for a few seconds without seeing
anything. Then her hands were gently taken, and she found herself
pushed into a low chair.
“Bring her some strawberries, Mr. Bonnington,” said Mrs. Dale. “By
the by, I may as well remark that I don’t intend to call you Mr.
Bonnington very long. I shall drop into plain ‘Rudolph’ very soon, if
only to give a fresh shock to the neighborhood, to whom I am
Shocker in ordinary.”
“The sooner the better. I can’t understand anybody’s being Mr.
Bonnington but my father. Now he looks equal to the dignity, while I
don’t. I always feel that there is a syllable too many for me, and that
people despise me in consequence.”
Mabin, who had recovered the use of her eyes, felt rather envious.
The quick give and take of light talk like this was so different from the
solemn conversations carried on at home, where her father laid
down the law and everybody else agreed with him, that she felt this
levity, while pleasant and amusing, to be something which would
have caused the good folk at home to look askance.
“And how have you been, child, since that unhappy day when I
saw you last?”
And Mrs. Dale came to the next chair, and piled sugar on the
strawberries.
 “Oh, I’ve been getting on all right, but it is tiresome not to be able
to walk without those things. And it has made me in everybody’s
way,” sighed Mabin.
“How is that?”
“Papa could have let the house to go abroad, as he wanted to,
when the accident happened. Only I couldn’t well be moved then.
And now that I could go, he has lost the house he had heard of at
Geneva, and one which he could have now is too small for us. So
that I feel I am in the way again.”
“Do you mean,” asked Mrs. Dale quite eagerly, “that they could go
if only they could dispose of you?”
“Yes. There is one room short.”
Little Mrs. Dale sprang up, and the color in her cheeks grew
pinker.
“Do you think,” she asked, after a moment’s pause, “that your
parents would allow you to stay with me? If you would come?” she
finished with a plaintive note of entreaty on the last words.
“Oh, I am sure they would, and I am sure I would!” cried Mabin,
with undisguised delight.
And then quite suddenly the face of the black-robed lady grew
ashy gray, and she sank down into her chair trembling from head to
foot.
“No. I—I mustn’t ask you,” she said hoarsely.
And there was a silence, during which both her young hearers cast
down their eyes, feeling that they dared not look at her. It was Mabin
who spoke first. Putting her hand between the two white hands of
Mrs. Dale, she said gently:
“Is it because you are lonely you want me to come?”
She did venture to look up then, startled by the shiver which
convulsed Mrs. Dale as she spoke. And in the blue eyes she saw a
look of terror which she never forgot.
“Lonely! Oh, child, you will never know how lonely!” burst from her
pale lips.
“Then I will come,” said Mabin. “I should like to come.”
There was another silence. Mrs. Dale had evidently to put strong
constraint upon herself to check an outburst of emotional gratitude.
Rudolph, moved himself by the little scene, was looking out of the
window. The lady in black presently spoke again, very gravely:
“I don’t think you will be very much bored, dear, and you will be
doing a great kindness to a fellow-creature. And yet—I hardly like—I
don’t feel that I ought——”
“But I feel that I must and shall,” said Mabin brightly. “You don’t
know how beautiful it would be for me to feel that at last, for a little
while, I shouldn’t be in the way!”
And the overgrown girl, who was snubbed at home, had tears in
her eyes at the remembrance of the kind touch which she had felt on
the day of her accident. Mrs. Dale was too much moved to say much
more, but it was agreed between the ladies that the suggestion
should be formally made by the tenant of “The Towers” to the heads
of the household at “Stone House” without delay, that Mabin should
stay with her new friend during the absence abroad of the rest of the
Rose family.
Mabin did notice, while they talked, that Rudolph remained not
only silent but somewhat constrained; but it was not until she took
her leave of Mrs. Dale, and he followed her out, that the young girl
attached any importance to his reserve.
Once out of hearing of Mrs. Dale, who stood on the stone steps to
bid them good-by, Rudolph asked her abruptly:
“Do you think they’ll let you come?”
“Oh, yes, they’ll only be too glad to get rid of me. Why do you ask
in that tone?”
“Well, there is something I think I ought to tell you, if you are
thinking of staying with Mrs. Dale.”
“Well, what is it?”
“It is that she is being watched.”
“Watched!”
“Yes, by a stranger, a man whom I have never seen in the place
before. He hovers about this place, keeping out of range of possible
eyes in the house, at all hours of the day and even of the night.”
“But how do you know this?”
The words slipped out of her mouth, and it was not until she saw
Rudolph redden that she saw that she was too inquisitive.
“I am sure of what I say, anyhow,” said he quietly.
 Mabin looked thoughtful.
“I don’t care!” she said at last.
“I thought you wouldn’t.”
“And I shan’t tell anybody anything about it.”
“I was sure you wouldn’t.”
“But I shall tell Mrs. Dale.”
Rudolph stopped and looked at her.
“I think you had better not do that,” he said.
“But why should a person watch her, except with the intention of
trying to do her some harm?”
“Well, I don’t know. But I think if you do tell her, knowing how
highly nervous she is, you will do her more harm than ever the
mysterious watcher would. Perhaps you would even drive her out of
the place, in which case most assuredly the watcher would go after
her, while if we keep her here perhaps we may manage to draw his
fangs.”
Mabin felt frightened. Then, being a matter-of-fact girl, she got the
better of this feeling quickly, and looked up keenly at her companion.
“What do you exactly mean by that?” she asked.
“Only that I will get hold of the man quietly and find out what his
little game is. Though I can guess.”
“Well, you can tell me what your guess is?”
“Why, debt, of course. One can see she is inclined to be
extravagant, and very likely she has run up bills somewhere. Don’t
you think that seems likely?”
His tone was rather anxious, Mabin thought. But she answered
indignantly:
“No, I don’t. It would be very dishonorable to run away without
paying one’s debts, and I don’t think you much of a friend to poor
Mrs. Dale to suggest such a thing!”
Rudolph looked not guilty, but grave.
“Well” said he, “people don’t hang about a place, at the risk of
getting taken up ‘on suspicion of loitering, for the purpose of
committing a felony,’ without some reason.”
“Why,” cried Mabin triumphantly, “that is the reason! Mrs. Dale has
some lovely diamond rings, and the loitering gentleman wants to
steal them!”
 “Well, perhaps you are right,” said Rudolph doubtfully.
“I am sure of it!” retorted Mabin resolutely. And she held out her
hand. “Good-by, and thank you for your help.”
“And you will remember my parable about the Roses?” said he, as
he took her hand and thought he liked gray eyes after all better than
blue ones.
“Perhaps,” said Mabin cautiously, as she hopped away on her
crutches.
 CHAPTER IV.
WAS IT A RECOGNITION?

While Mabin was still talking to Rudolph in the road between “The
Towers” and “Stone House,” a tall parlormaid, in snow-white French
cap and ends, passed them, on her way from the former to the latter
house, bearing a letter in her hand.
And when Mabin reached home, she found that the Powers had
already received Mrs. Dale’s invitation to Mabin.
In truth it had put both husband and wife into a position of some
difficulty. For while, on the one hand, they were delighted at this
opportunity of getting “the one too many” off their hands for a time,
yet there were the opinions of their neighbors to be considered; and
the tide of public feeling had set in strongly against the lady in black.
If her hair had been dark instead of fair, it would have made all the
difference. The beauty which goes with brown hair and a more or
less dark complexion is not so startling, not so sensational, as the
loveliness of pink and white and gold which made Mrs. Dale so
conspicuous. If again, she had not been in mourning, and such
pretty mourning, they would have been readier to make allowance
for her eccentricities. But the knowing ones had begun to discover
that there were discrepancies in her attire, that her mourning was
either too deep for diamond rings to be permissible, or not deep
enough for the heavy black veil she wore.
So that, in short, it was now almost universally admitted that this
person with the too showy carriage and horses, and the dangerously
pretty face, was an individual to be avoided, and it was decided that
her reluctance to enter the best society of the place, when that
society had held out its uninviting arms to her, arose from a
wholesome fear that the wise women of the place would “find her
out.”
Mr. Rose read Mrs. Dale’s note twice through, very slowly, as if
trying to discover hidden meanings in its simple words. Then he
looked at his wife, who was watching him rather anxiously.
 “Well, my dear, and what do you think?” asked he.
It pleased him to ask her opinion thus on most things, not that he
ever had any intention of heeding her wishes in preference to his
own, but in the hope that she would express some modest inclination
one way or the other, to give him an impetus in the opposite
direction.
“I think, dear, that it would hardly do,” murmured the lady, hoping
that for once her husband would fall in with her views. “You must
have heard the way in which people talk about this Mrs. Dale, so that
it would be thought very strange if we let Mabin stay with her. Don’t
you think it was rather underhand of her to get hold of the child this
afternoon?”
“Underhand! Certainly not,” replied Mr. Rose with decision. “The
most natural thing in the world, considering how kind she was to the
girl at the time of her accident. And as for the talk of the place, why, if
you listened to all the old women say you would never go outside
your door for fear your neighbors should think you were going to
steal their hens!”
There was a pause. She would not irritate him by another remark.
So he presently went on:
“I suppose you think the Vicar’s wife would scold you?”
“Not scold, of course, but I am sure she would disapprove,” said
Mrs. Rose meekly.
“Ah! I thought so. Well, I will give the old lady something to talk
about then. Mabin shall stay with Mrs. Dale if she wishes.”
Mrs. Rose sighed heavily.
“She will wish it, of course. Girls always wish to do the very things
which are not proper for them.”
“You may be quite satisfied, Emily, that what I allow my daughter
to do is quite proper,” said Mr. Rose severely, as he left the room.
Mrs. Rose sighed. She had not told him, because it would have
been of no use, that she had to be more particular than he about
Mabin, because, being the girl’s step-mother only, she was the more
exposed to the gossip of the neighborhood—a force she dreaded—
than her husband was. But she vented her ill-humor on Mabin
herself, whom she informed very acidly that if she chose to go to
“this Mrs. Dale,” and was not comfortable with her, the fault was hers
and her father’s.
Mabin received these remarks meekly, rejoicing in the
approaching holiday. She had nothing very serious to complain of in
her treatment under her father’s roof, but the snubs of her father, the
tacit dislike of her step-mother, and the fact that the difference in age
between her half-sisters and herself left her much alone, all
combined to make her welcome the change.
Emily and Ethel, who were fourteen and twelve years of age,
insipid and spiritless young persons with little brown eyes and little
brown pigtails, teased her with questions about her visit of the
afternoon.
“Is it true that her hair’s dyed?” asked Emily, getting Mabin into a
corner after tea.
“No, of course it’s not,” was the indignant answer.
“Oh, well, you needn’t be so angry. Miss Bradley said she was
sure it was, and that she knew the very stuff she used.”
“Miss Bradley had better try the stuff on her own wisp then,”
retorted Mabin angrily.
“What is the house like inside, Mabin?” asked Ethel, who, though
only twelve, was quite as great a gossip as there was in the parish.
“Why, there were chairs and tables in it, just as there are in every
other house. What do you suppose it was like?”
“Mabin, don’t snub the children. Their interest is very natural,” said
Mr. Rose peevishly from the other end of the room.
“Horsehair and mahogany, red moreen curtains, and a black
marble clock on the mantelpiece,” said Mabin laconically.
“Why, that doesn’t sound very nice, that you should be in such a
hurry to go there!” objected Ethel. “But perhaps the other rooms are
better.”
“Very likely,” said Mabin.
But Mabin was really just a little bit alarmed at her own good
fortune in getting her father’s consent so easily. She had a
superstitious feeling, in spite of her reputed strength of mind, that
anything worth having ought to be rather difficult to get. In spite of all
her loyalty to her heroine, too, she thought more often than she
wished about Rudolph’s ridiculous fancy that Mrs. Dale was
watched. And although she always dismissed the thought by saying
to herself that Rudolph was in love with the lovely widow, and
therefore “fancied things,” she was anxious to meet him again and to
learn whether he still thought that the fair tenant of “The Towers” was
being watched.
In the mean time great confusion reigned at “Stone House.”
Everybody was immersed in the horrors of “packing up,” and it
was impossible to go upstairs without encountering people
staggering under the burden of a heap of things which would have
been better left behind. Even the authority of Mr. Rose, who disliked
the daily routine to be disturbed, failed to get any meal eaten at the
proper time, or without unnecessary hurry.
Even the fact that Mr. Rose’s old friend Mrs. Haybrow was
expected on a short visit before the migration, failed to check the fury
of the packers. It was unfortunate that she should come at such a
time, certainly. But Mrs. Rose reckoned on inflaming her friend’s
mind with her own zeal, and inwardly proposed orgies of competitive
trunk-filling to while away the visitor’s time.
Mrs. Rose secretly hoped, too, that Mrs. Haybrow, through her
connections at Todcaster, would be able to furnish her husband with
proofs that Mrs. Dale was not a person to be encouraged. It was not
yet too late to put off Mrs. Dale, although Mr. Rose had called upon
that person to thank her for the invitation to his daughter and to
accept it.
It was the day of Mrs. Haybrow’s expected arrival, and Mr. and
Mrs. Rose had driven to the station to meet her. Mabin, wondering
whether the visitor, whom she had not seen since she was a child,
would be “nice,” was hobbling along the garden path, rather painfully
indeed, but at last without a crutch, when she heard a great rustling
of the branches of the lilac bushes which grew close under the wall.
And then, above the wall, she saw the face of Rudolph.
“Oh!” cried Mabin, with a little fluttering of the heart. “I—I thought
you had gone back to your ship!”
“Why, so I had,” replied Rudolph, raising himself so that she had a
view of his shoulders as well as of his head. “But I’ve come back
again, you see.”
 “I can guess what brought you!” said Mabin, wishing the next
moment that she had not uttered the words.
Rudolph took her up quickly.
“Can you? Well then, what was it?”
Mabin blushed scarlet. Of course the thought that was in her mind
that the charms of the fascinating widow had drawn him to Stone.
And just the least little twinge of nascent jealousy had given a sting
of pique to her tone. But she would not for the world have owned to
this; and the mere thought that he might have guessed it was misery.
As she did not answer, Rudolph shook his head.
“I don’t think you are quite as clever as you suppose,” he said.
“And I don’t choose to tell you what brought me back. But I may just
warn you that you are likely to get tired of the sight of me before I do
go away altogether, as I can get as much leave as I like while my
ship is at Portsmouth. Rather alarming prospect, isn’t it?”
“You will get tired of being on shore, won’t you?” asked Mabin, not
feeling equal to answering him in his own tone, which was what her
parents and the Vicarage people would have called “flippant.”
“That depends,” said Rudolph, looking down with interest at the
dried-up blossoms of the lilac trees.
Mabin glanced at him, and began to hope nervously that she might
not see too much of him. She had never seen a man whom she
considered so handsome as this brown-faced young lieutenant with
the merry black eyes, who made her feel so ridiculously shy and stiff.
And the very attraction he had for her seemed to the simple young
girl alarming, since she raised him in her maiden fancy to a pinnacle
from which such a peerless creature could never descend to her.
In spite of herself her tone sounded cold and constrained,
therefore, as she cut short the pause in the conversation by asking if
they were all well at the Vicarage.
“Quite well, thank you,” answered Rudolph demurely. “I suppose
that kind inquiry is meant for a snub, isn’t it? And intended to imply
that I ought not to have addressed you in this informal manner over
the wall, but that I ought to have called in the proper manner at the
other side of the house?”
“It wasn’t meant to imply that,” replied Mabin with solemn
straightforwardness. “Only I wanted to say something, and I had
nothing better to say. I must tell you that everybody says that I have
no conversation.”
“People allow you very few good points, according to your
account.”
“Quite as many as I have, though!”
“Well, at least, then you have one merit of unusual modesty.”
Mabin looked up at him steadily and frankly.
“It’s rather difficult for me to talk to you, because I can never tell
whether you are serious or only laughing at me. Don’t you rather—
rather puzzle Mrs. Bonnington?”
“Well, I am afraid I rather—rather shock her too.”
“We must all seem, down here, very antediluvian to you. There is
only one person about here you can speak to in your own way.”
“Mrs. Dale?”
To Mabin’s fanciful and rather jealous eyes it seemed that
Rudolph’s color grew a little deeper as he uttered the name.
“Yes.”
“Ah! You will have an opportunity of learning the art, if you are
going to stay with her.”
“But it is an art which will be entirely useless when I get back here
again. Papa and mamma would think it rather shocking. Do you
know, if they knew how lively Mrs. Dale is in her ordinary talk they
wouldn’t let me go to her?”
“Then I shouldn’t tell them, if I were you. You will find a use for the
art of conversation some day, you know, when you come across
other frivolous and good-for-nothing young persons, like Mrs. Dale
and me.”
Mabin would rather he had not coupled his name with that of the
lovely widow.
“Were Mr. and Mrs. Bonnington interested to hear you had been to
see her?” asked Mabin, feeling as she spoke that this was another
indiscretion.
But Rudolph began to laugh mischievously.
“They would have been extremely ‘interested,’ I am sure, if they
had heard of it,” said he. “But I have too much consideration for my
parents to impart to them any information which would ‘interest’ them
too deeply to be good for their digestions. I suppose you think that
shocking, don’t you?”
But Mabin was cautious. There was more than one gulf, she felt,
between her and the merry young sailor, and she was not going to
make them any wider.
“I’m sure you do what is best,” she said modestly.
“I’ve got something to tell you,” said he. “But it’s rather a
confidential communication, and these lilac bushes extend a long
way. Will you come nearer to the wall? Or may I get over it?”
“You may get over if you like,” said Mabin, coming as she spoke a
little nearer to the bushes.
Rudolph availed himself of the permission in the twinkling of an
eye, and stood beside her on the grass path under the limes, looking
down at the pretty nape of the girlish neck, which showed between
the soft brown hair and the plain, wide turn-down collar of pale blue
linen which she wore with her fresh Holland frock.
“The man—I told you about the man I saw watching ‘The Towers,’
well, he has disappeared,” said Rudolph, not sorry to have an
excuse for whispering into such a pretty little pink ear.
“Oh! I am glad!”
“So I hope we shall see no more of him.”
“And do you still think—surely you can’t still think—that he was
watching Mrs. Dale?”
“Oh, well, don’t know, of course. And at any rate the slight
objection I had to your going to ‘The Towers’ has disappeared.”
Mabin felt a strange pleasure at the interest implied in this concern
for her. There was a pause, broken by Mabin, who suddenly started,
as if waking from a dream.
“The carriage!” cried she. “They have come back. I must go in.
Good-by.”
She held out her hand. He took it, and detained it a moment.
“I may come and see you sometimes, when you are at ‘The
Towers,’ mayn’t I? For old acquaintance’ sake?”
“Or—for Mrs. Dale’s!” said Mabin quickly, as she snatched away
her hand and ran into the house.
She was not so silly as not to know where the attraction of ‘The
Towers’ lay!
 Mrs. Rose’s lumbering old landau, which made such a contrast to
Mrs. Dale’s smart victoria, had returned from the station, and as it
drove slowly along the road past ‘The Towers,’ Mrs. Rose was just
finishing to Mrs. Haybrow a long recital of her difficulties in
connection with the doubtful new resident.
Mr. Rose had chosen to come back on foot, so his wife could pour
out her tale without interruption or contradiction.
“There,” she cried below her breath, as they came close to the
gates of ‘The Towers,’ “you will be able to see her. She is standing
just inside the garden, calling to her little dog. Don’t you think that a
little dog always looks rather—rather odd?”
Mrs. Haybrow thought that this was somewhat severe judgment,
but she did not say so. She got a good view of the mysterious lady in
black; for Mrs. Dale raised her golden head as the carriage passed,
and she and Mrs. Rose exchanged a rather cool bow.
To the great surprise of her companion, Mrs. Haybrow fell
suddenly into a state of intense excitement.
“Why, it’s Dolly Leatham, little Dolly Leatham!” she cried with
evident delight. “The idea of my meeting her down here! I haven’t
seen the child for years.”
“You know her then?” asked Mrs. Rose, in a tone which in relief
was mingled with disappointment at the collapse of her own
suspicions.
“I used to know her very well. She was the belle of that part of
Yorkshire. The last I heard of her was that she was engaged to be
married to some man who had a lot of money; and they said she was
being hurried into it by her people rather against her will.”
“Well, she has managed to get rid of him,” said Mrs. Rose coldly.
“You see she is in widow’s dress now.”
“Yes, so I see. Poor Dolly! It seems rather strange to find her here,
so far from all her friends! And the things you have told me.”
After a pause Mrs. Haybrow said decidedly: “I must call upon her
to-morrow—No, I’ll go and see her at once. There will be plenty of
time before dinner, won’t there? There’s something mysterious about
this, and I must find out what it is.”
So, when she had had a cup of tea, Mrs. Haybrow went straight to
“The Towers.”