Deeply learned face representations are sparse, selective, and robust

Slides:



Advertisements
Similar presentations
Face Recognition Sumitha Balasuriya.
Advertisements

Active Appearance Models
Face Recognition: A Convolutional Neural Network Approach
Aggregating local image descriptors into compact codes
Zhimin CaoThe Chinese University of Hong Kong Qi YinITCS, Tsinghua University Xiaoou TangShenzhen Institutes of Advanced Technology Chinese Academy of.
Face Recognition. Introduction Why we are interested in face recognition? Why we are interested in face recognition? Passport control at terminals in.
Proposed concepts illustrated well on sets of face images extracted from video: Face texture and surface are smooth, constraining them to a manifold Recognition.
Face Representations Learning by Deep Models
Face Recognition and Biometric Systems
Large-Scale, Real-World Face Recognition in Movie Trailers Week 2-3 Alan Wright (Facial Recog. pictures taken from Enrique Gortez)
Face Recognition Under Varying Illumination Erald VUÇINI Vienna University of Technology Muhittin GÖKMEN Istanbul Technical University Eduard GRÖLLER Vienna.
Principal Component Analysis
Graz University of Technology, AUSTRIA Institute for Computer Graphics and Vision Fast Visual Object Identification and Categorization Michael Grabner,
CONTENT BASED FACE RECOGNITION Ankur Jain 01D05007 Pranshu Sharma Prashant Baronia 01D05005 Swapnil Zarekar 01D05001 Under the guidance of Prof.
Computer Vision I Instructor: Prof. Ko Nishino. Today How do we recognize objects in images?
Smart Traveller with Visual Translator for OCR and Face Recognition LYU0203 FYP.
Methods in Leading Face Verification Algorithms
Face Recognition: An Introduction
Oral Defense by Sunny Tang 15 Aug 2003
Face Detection and Recognition
Face Recognition Using Neural Networks Presented By: Hadis Mohseni Leila Taghavi Atefeh Mirsafian.
Face Recognition Using EigenFaces Presentation by: Zia Ahmed Shaikh (P/IT/2K15/07) Authors: Matthew A. Turk and Alex P. Pentland Vision and Modeling Group,
Jinhui Tang †, Shuicheng Yan †, Richang Hong †, Guo-Jun Qi ‡, Tat-Seng Chua † † National University of Singapore ‡ University of Illinois at Urbana-Champaign.
Multiclass object recognition
Presented by: Kamakhaya Argulewar Guided by: Prof. Shweta V. Jain
BDPCA Plus LDA: A Novel Fast Feature Extraction Technique for Face Recognition 授課教授 : 連震杰 老師 組員 : 黃彥綸 何域禎 W. Zuo, D. Zhang, J. Yang, K. Wang, “BBPCA plus.
嵌入式視覺 Pattern Recognition for Embedded Vision Template matching Statistical / Structural Pattern Recognition Neural networks.
Person-Specific Domain Adaptation with Applications to Heterogeneous Face Recognition (HFR) Presenter: Yao-Hung Tsai Dept. of Electrical Engineering, NTU.
Eigenedginess vs. Eigenhill, Eigenface and Eigenedge by S. Ramesh, S. Palanivel, Sukhendu Das and B. Yegnanarayana Department of Computer Science and Engineering.
Mining Discriminative Components With Low-Rank and Sparsity Constraints for Face Recognition Qiang Zhang, Baoxin Li Computer Science and Engineering Arizona.
Introduction to Computer Vision Olac Fuentes Computer Science Department University of Texas at El Paso El Paso, TX, U.S.A.
Graph Embedding: A General Framework for Dimensionality Reduction Dong XU School of Computer Engineering Nanyang Technological University
Access Control Via Face Recognition Progress Review.
Local Non-Negative Matrix Factorization as a Visual Representation Tao Feng, Stan Z. Li, Heung-Yeung Shum, HongJiang Zhang 2002 IEEE Presenter : 張庭豪.
IEEE TRANSSACTIONS ON PATTERN ANALYSIS AND MACHINE INTELLIGENCE
Using Support Vector Machines to Enhance the Performance of Bayesian Face Recognition IEEE Transaction on Information Forensics and Security Zhifeng Li,
A Two-level Pose Estimation Framework Using Majority Voting of Gabor Wavelets and Bunch Graph Analysis J. Wu, J. M. Pedersen, D. Putthividhya, D. Norgaard,
Pattern Recognition April 19, 2007 Suggested Reading: Horn Chapter 14.
Face Recognition: An Introduction
CSE 185 Introduction to Computer Vision Face Recognition.
A NOVEL METHOD FOR COLOR FACE RECOGNITION USING KNN CLASSIFIER
Face Image-Based Gender Recognition Using Complex-Valued Neural Network Instructor :Dr. Dong-Chul Kim Indrani Gorripati.
2D-LDA: A statistical linear discriminant analysis for image matrix
Face detection and recognition Many slides adapted from K. Grauman and D. Lowe.
Computer Vision Lecture 7 Classifiers. Computer Vision, Lecture 6 Oleh Tretiak © 2005Slide 1 This Lecture Bayesian decision theory (22.1, 22.2) –General.
WLD: A Robust Local Image Descriptor Jie Chen, Shiguang Shan, Chu He, Guoying Zhao, Matti Pietikäinen, Xilin Chen, Wen Gao 报告人:蒲薇榄.
Martina Uray Heinz Mayer Joanneum Research Graz Institute of Digital Image Processing Horst Bischof Graz University of Technology Institute for Computer.
Deep Learning Overview Sources: workshop-tutorial-final.pdf
Facial Smile Detection Based on Deep Learning Features Authors: Kaihao Zhang, Yongzhen Huang, Hong Wu and Liang Wang Center for Research on Intelligent.
Big data classification using neural network
Hybrid Deep Learning for Reflectance Confocal Microscopy Skin Images
A Discriminative Feature Learning Approach for Deep Face Recognition
Deep Learning Amin Sobhani.
University of Ioannina
Unsupervised Riemannian Clustering of Probability Density Functions
Hybrid Features based Gender Classification
Recovery from Occlusion in Deep Feature Space for Face Recognition
Final Year Project Presentation --- Magic Paint Face
Cheng-Ming Huang, Wen-Hung Liao Department of Computer Science
State-of-the-art face recognition systems
Outline Peter N. Belhumeur, Joao P. Hespanha, and David J. Kriegman, “Eigenfaces vs. Fisherfaces: Recognition Using Class Specific Linear Projection,”
By: Kevin Yu Ph.D. in Computer Engineering
Deep Face Recognition Omkar M. Parkhi Andrea Vedaldi Andrew Zisserman
“Bayesian Identity Clustering”
A Proposal Defense On Deep Residual Network For Face Recognition Presented By SAGAR MISHRA MECE
Age and Gender Classification using Convolutional Neural Networks
John H.L. Hansen & Taufiq Al Babba Hasan
Paper Reading Dalong Du April.08, 2011.
Face Recognition: A Convolutional Neural Network Approach
Goodfellow: Chapter 14 Autoencoders
Presentation transcript:

Deeply learned face representations are sparse, selective, and robust Authors: Yi Sun1 Xiaogang Wang2;3 Xiaoou Tang 1;3 1Department of Information Engineering, The Chinese University of Hong Kong 2Department of Electronic Engineering, The Chinese University of Hong Kong 3Shenzhen Institutes of Advanced Technology, Chinese Academy of Sciences Course:CSC8860 Seminar on Computer Vision and Pattern Recognition Instructor: Ming Dong Presenter: Lu Wang

Contents Introduction DeepID2 Methods Comparison of DeepID2 and DeepID2+ Performance Conclusion

Introduction-LDA Each face is represented by a large number of pixel values. Linear discriminant analysis is primarily used here to reduce the number of features to a more manageable number before classification. Each of the new dimensions is a linear combination of pixel values, which form a template. The linear combinations obtained using Fisher's linear discriminant are called Fisher faces, while those obtained using the related principal component analysis are called eigenfaces.

Introduction- Bayesian face Direct visual matching of images for the purposes of face recognition and image retrieval, using a probabilistic measure of similarity, based primarily on a Bayesian (MAP) analysis of image differences. The performance advantage of this probabilistic matching technique over standard Euclidean nearest 􏱒neighbor eigenface matching.

Introduction- Unified Subspace Decomposes face difference into three components, intrinsic difference, transformation difference, and noise. Use the face difference model and a detailed subspace analysis on the three components to develop a unified framework for subspace analysis. Construct of a 3D parameter space that uses three subspace dimensions as axis.

Introduction- DeepID2(Deep IDentification-verification features) Problem/Challenge Develop effective feature representations for reducing intra-personal variations while enlarging inter-personal differences. Solution Deep learning with using both face identification and verification signals as supervision.

Introduction- DeepID2 Source data LFW (Labeled Faces in the Wild):a database of face photographs designed for studying the problem of unconstrained face recognition. The data set contains more than 13,000 images of faces collected from the web. Each face has been labeled with the name of the person pictured. 1680 of the people pictured have two or more distinct photos in the data set. The only constraint on these faces is that they were detected by the Viola-Jones face detector.

Introduction- DeepID2 Identification-verification guided deep feature learning

Introduction- DeepID2

Introduction- DeepID2

Introduction- DeepID2 Face Verification SDM algorithm to detect 21 facial landmarks Joint Bayesian model for face verification based on the extracted DeepID2:

Introduction- DeepID2 Forward-backward greedy algorithm to select a small number of effective and complementary DeepID2 vectors (25 in experiment), which saves most of the feature extraction time during test.

Introduction- DeepID2 Balancing the identification and verification signals

Introduction- DeepID2 Balancing the identification and verification signals

Introduction- DeepID2 Rich identity information improves feature learning

Introduction- DeepID2 Investigating the verification signals

Introduction- DeepID2 Final system and comparison with other methods

Methods-Comparison of DeepID2 and DeepID2+ DeepID2+ nets are inherited from DeepID2 nets, which have four convolutional layers, with 20 , 40 , 60 ,and 80 feature maps, followed by a 160 -dimensional feature layer fully-connected to both the third and fourth convolutional layers. The 160 -dimensional feature layer (DeepID2 feature layer) is supervised by both face identification and verification tasks.

Methods-Comparison of DeepID2 and DeepID2+ Compared to DeepID2, DeepID2+ makes three improvements. First, it is larger with 128 feature maps in each of the four convolutional layers. The final feature representation is increased from 160 to 512 dimensions. Second, our training data is enlarged by merging the CelebFaces+ dataset,the WDRef dataset [6], and some newly collected identities exclusive from LFW. The larger DeepID2+ net is trained with around 290; 000 face images from 12; 000 identities compared to 160; 000 images from 8; 000 identities used to train the DeepID2 net. Third, in the DeepID2 net, supervisory signals are only added to one fully-connected layer connected to the third and fourth convolutional layers, while the lower convolutional layers can only get supervision with gradients back-propagated from higher layers.

Performance High-performance of DeepID2+ nets

Performance High-performance of DeepID2+ nets

Performance High-performance of DeepID2+ nets

Performance Moderate sparsity of neural activations

Performance Neural activation distribution

Performance Neural activation distribution

Performance Neural activation distribution

Performance Neural activation distribution

Performance Robustness of DeepID2+ features

Performance Robustness of DeepID2+ features

Performance Robustness of DeepID2+ features

Conclusion This paper designs a high-performance DeepID2+ net which sets new sate-of-the-art on LFW and YouTube Faces for both face identification and verification. This work not only significantly advances the face recognition performance, but also provides valuable insight to help people to understand deep learning and its connection with many existing computer vision researches such as sparse representation, attribute learning and occlusion handling.

Thank you for listening!