You are on page 1of 12

COMPARING FACIAL

RECONGITION TECHNQUES
Computer Vision and its
application
o Interdisciplinary scientific field that deals with images and videos
to automate human visual tasks.
o Involves machine learning ,deep learning ,Artificial intelligence,
computer graphics and image processing
o Computer vision applications include self-driving cars, terrain
detection using drones, Augmented reality (AR),3D scanning
o Object detection is a sub-domain of computer vision
o Object detection deals with detecting instances of semantic objects
o Research domain of object detection includes facial detection
o Object detection applications includes video survilleance,facial
recognition.
Facial Recognition system
o Identifying one or more faces in image or video using a database of
stored face images
o Face detection is first step to face recognition
o Requires both location of face in image (position) and the extent face
is localized (within bounding box)
o Face detection dived into two main groups, Feature-based and
image-based
o Feature-based detection uses filters that search and locate faces,
while image based detection learns to automatically locate and
extract face from entire image
o Applications of facial recognition includes surveillance, attendance
tracking, identity validation system
Learning Affective Video Features for Facial Expression
Recognition via Hybrid Deep Learning 1
(March,2019),IEEE Access
 A challenging issue in Facial expression Recognition (FER) is to extract spatiotemporal video
features from video sequence
 New proposed method of FER in video sequences combine hybrid deep learning model
 First CNN employs spatial CNN and a temporal CN network to separately learn high-level
spatial and temporal features on divided video segments. These algorithms are fined tuned on
target video datasets from pre-trained CNN model.
 Obtained features are integrated into deep-fusion network built with DBN (Deep Belief
Network) model
 Then , a linear Support Vector machines (SVM) is adopted to perform facial expression
classification tasks.
 So, four steps were performed.
• Generation of CNN inputs
• Spatio-temporal learning with CNNs
• Spatio-temporal fusion with DBNs
• Video based expression classification

SHIQING ZHANG 1, XIANZHANG PAN1, YUELI CUI1, XIAOMING ZHAO1, AND LIMEI LIU2 1Institute of
Intelligent Information Processing, Taizhou University, Taizhou 318000, China 2Institute of Big Data and
Internet Innovation, Hunan University of Commerce, Changsha 410205, China
Results
A Face Emotion Recognition Method Using
Convolutional Neural Network and Image Edge
Computing (October ,2019), IEEE access
2
 First data preprocessing is carried out before images are feed for training
 Haar-classifier is trained using haar-like features combined with Adaboost algorithm.
 This combination would extract facial features and screens out important features
from haar-like features.
 The features are fed to the convolution network. Edges of each layer of the image is
extracted in convolution.
 Extracted information is superimposed on each feature image to preserve the
texture of image.
 Dimensionally reduction of features carried out in the pooling process
 Finally, test image is classified and recognized by the softmax classifier
 For verification, Feb 2013 database of facial expression is simulated with LFW
dataset

HONGLI ZHANG 1, ALIREZA JOLFAEI2, AND MAMOUN ALAZAB 3 1Department of Educational Technology, Inner Mongolia
Normal University, Hohhot 010022, China 2Department of Computing, Macquarie University, Sydney, NSW 2109,
Australia 3Charles Darwin University, Darwin, NT 0810, Australia
RESULTS
Dictionary Representation of Deep Features for Occlusion-
Robust Face Recognition (February, 2019) IEEE access
3
 Deep learning achieved exciting results in face recognition ,however it struggled with
occluded faces
 Deep dictionary Representation-based classification (DDRC) method is proposed
 CNN is used as a feature extractor, followed by a dictionary to linear code the extracted
deep features
 Dictionary is composed of deep features of training samples and auxiliary part consisting of
mapping vectors acquired from subject from inside or outside the training set and
associated with occlusions patterns of testing face samples.
 Proposed scheme generic for both occluded, non-occluded faces.
 VGG-face network is adopted as the deep feature extractor for proposed DDRC because of
public availability and demonstrated face verification accuracy close to human on LFW
dataset.
 Four type of auxiliary dictionaries consisting of pair matching with sunglasses and scarf
patterns.

FENG CEN 1 AND GUANGHUI WANG 2, (Senior Member, IEEE) 1Department of Control Science and
Engineering, College of Electronics and Information Engineering, Tongji University, Shanghai 201804,
China 2Department of Electrical Engineering and Computer Science, The University of Kansas, Lawrence,
KS 66045, USA
RESULTS
RESULTS
ANALYSIS
S. Title Dataset Methodology Conclusion Future proposal
No
01 Learning Affective
Video Features for
BAUM-1s,
RML,
Two CNN including
spatial and temporal
VGG16 pretrained
imagedata fed to learn
Exploring deep
compression of deep
Facial Expression MMI CNN. Results are spatiotemporal models to reduce the
Recognition via inducted to DBN.SVM features. Result are network parameters of
Hybrid Deep employed for facial tested on public deep models.
Learning classification task dataset.
(March,2019),IEEE
Access

02 A Face Emotion
Recognition Method
Fer-2013
facial
CNN and image edge
detection
Compared to RCNN and
FR-CNN, convergence
To deal with noise
issues,i.e face posture,
Using Convolutional expression speed of proposed occlusions more robust
Neural Network and database model is much faster in models will be
Image Edge complex background suggested. Focus on
Computing how to reduce
(October ,2109), complexity of network
IEEE access structure, try to
recognize dynamic
expressions with 3d
convolution

03 Dictionary
Representation of
VGG
Facemask
CNN as feature
extractor
Proposed DDRC is
generic for both
Occlusion pattern is
included in the auxiliary
Deep Features for with occlusion and non- dictionary. We will built
Occlusion-Robust configuratio occlusion face images. auxiliary library that can
Face Recognition n A[40] In addition it can handle handle occlusions
(February, 2019) , AR one sample per subject outside auxiliary
IEEE access database for simple application dictionary.
scenarios.

You might also like