Project topics and materials
Get Complete Project Material File(s) Now! »
Basic Tools and Concepts in Face Biometrics
Face recognition [159], the least intrusive biometric technique from the acquisition point of view, has been applied to a wide range of commercial and law enforcement applications. In comparison to other popular biometric characteristics (such as fingerprint [81] and iris [49]), biometric recognition (for either identification or verification) using face requires less user cooperation and thus can be integrated in many advanced conditions (notably in video surveillance). In this section, we will give a brief review of general face recognition by in- troducing the following four parts: (1) standard techniques, (2) evaluation metrics, (3) challenges and (4) face database with occlusions. In the first part we will review the three most representative techniques, namely the Eigenface [140], the Fisherface [21] and the LBP based face recognition [17]. In part two, the identification mode and verification mode with their corresponding evaluation metrics are described. We then discuss the most often facial variations including expression, illumination, pose and occlusion in the third part. Part four gives the overview of the AR face database.
Standard Techniques
In this section, we briefly review three most widely used (and the mostly cited as well) face recognition techniques, namely Eigenface [140], Fisherface [21] and LBP based face recog- nition [17]. All those techniques are considered as the benchmarks for more advanced al- gorithm development. The good understanding of these methods is a key to design and develop occlusion-robust face recognition algorithms.
Review of Face Recognition under Occlusions
The classical methodology to address face recognition under occlusion is to find corruption- tolerant features or classifiers. Toward this goal, numerous previous works confirmed that locally emphasized algorithms are less sensitive to partial occlusions. Penev and Atick [122] proposed the local feature analysis (LFA) to extract local features by second order statistics. Martinez [107] proposed a probabilistic approach (AMM) which can compensate for par- tially occluded faces. Tan et al. [138] extended Martinez’s work by using the self-organizing map (SOM) to learn the subspace instead of using the mixture of Gaussians. In [89], Jim et al. proposed a method named locally salient ICA (LS-ICA) which only employs locally salient information in constructing ICA basis. In [62], Fidler et al. presented a method which com- bines the reconstructive and discriminative models by constructing a basis containing the complete discriminative information. Park et al. [120] proposed to use a line feature based face attributed relational graph (ARG) model which encodes the whole geometric structure information and local features of a face. Zhang et al. [157] proposed a non-statistical face representation – Local Gabor Binary Pattern Histogram Sequence (LGBPHS) to exploit the multi-resolution and multi-orientation Gabor decomposition. In [84], Jia and Martinez proposed the use of partial support vector machines (PSVM) in scenarios where occlusions may occur in both the training and testing sets. More recently, facial occlusion handling under the sparse representation based classification (SRC) [146] framework has demonstrated impressive performances in face recognition with occlusions. The idea of using SRC for occluded face recognition is first introduced byWright et al. [146], where an occluded face is represented as a linear combination of the whole face gallery added by a vector of errors (occlusion) in the pixel-level and the classification is achieved by l1 minimization. Zhou et al. [160] extend [146] by including a Markov Random Fields (MRF) model to enforce spatial continuity for the additive error vector to address contiguous occlusions. In [150], Yang and Zhang applied compressible image Gabor features instead of original image pixels as the feature vector used in SRC to reduce computations in the presence of occlusions. Liao and Jain [96] incorporated the SIFT descriptor into the SRC framework to achieve alignment free identification. Yang et al. [151] proposed a Robust Sparse Coding (RSC) method which seeks the maximum likelihood estimation (MLE) solution of the sparse coding problem for non-Gaussian/Laplacian occlusions in an iterative manner. Even though the SRC based methods achieve significant identification results on occluded faces from standard face databases (i.e. AR face database [106]), the prerequisite of those methods relies on the large number of training samples of each identity with sufficient variations. But in many practical face recognition scenarios, the training samples of each subject are often insufficient (the “curse of the dimensionality” [52] problem, in the extreme case only one template face per subject is available).
Lately, a few works have revealed that prior knowledge of occlusions can significantly improve the accuracy of local-feature/local-component based face recognition. Rama et al. [127] empirically showed that prior knowledge about occlusion (manually annotated) can improve Eigenface in local patches. In [116], Oh et al. have proposed an algorithm based on local non-negative matrix factorization (LNMF) [94], named Selective LNMF (S-LNMF) that automatically detects the presence of occlusion in local patches; face matching is then performed by selecting LNMF representation in the non-occluded patches. Zhang et al. [156] proposed to use Kullback-Leibler divergence (KLD) to estimate the probability distribution of occlusions in the feature space, so as to improve the standard LGBPHS based method [157] for partially occluded face.
Table of contents :
1 Introduction
1.1 Motivations
1.2 Content of the Thesis
1.3 Contributions
1.4 Outline
2 State-of-the-Art
2.1 Basic Tools and Concepts in Face Biometrics
2.1.1 Standard Techniques
2.1.2 Evaluation Metrics
2.1.3 Challenges
2.1.4 Face Database with Occlusions
2.2 Review of Face Recognition under Occlusions
2.3 State-of-the-Art Techniques
2.3.1 Locality Emphasized Algorithms
2.3.2 Sparse Representation based Classification (SRC)
2.3.3 Occlusion Analysis + Face Recognition
2.4 Conclusions
3 Classical Occlusion Handling for Robust Face Recognition
3.1 Introduction
3.2 Occlusion Analysis
3.2.1 Occlusion Detection in Local Patches
3.2.2 Occlusion Segmentation
3.3 Face Recognition
3.3.1 Improving LBP based Face Recognition
3.3.2 Improving LGBP based Face Recognition
3.4 Conclusions
4 Advanced Occlusion Handling for Robust Face Recognition
4.1 Introduction
4.2 Sparse Occlusion
4.2.1 Overview
4.2.2 Problem Statement
4.2.3 Method
4.2.4 Results
4.3 Dynamic Occlusion
4.3.1 Overview
4.3.2 Challenges and Innovations
4.3.3 Method
4.3.4 Results
4.4 Conclusions
5 Exploiting New Sensor for Robust Face Recognition
5.1 Introduction
5.2 Kinect Face Database
5.2.1 Overview
5.2.2 Review of 3D Face Database
5.2.3 KinectFaceDB
5.2.4 Experiments
5.3 Depth Assisted 2D Face Recognition Under Partial Occlusions
5.3.1 Overview
5.3.2 RGB based Occlusion Analysis to Improve Face Recognition
5.3.3 Depth based Occlusion Analysis to Improve Face Recognition
5.3.4 Results
5.4 Conclusions
6 Improving Baseline Face Recognition Methods
6.1 Introduction
6.2 Improving 2D Face Recognition via Combination of LBP and SRC
6.2.1 Overview
6.2.2 Background and Related Algorithms
6.2.3 Method
6.2.4 Results
6.3 Improving 3D Face Recognition via Multiple Intermediate Registration
6.3.1 Overview
6.3.2 The System
6.3.3 Results
6.4 Conclusions
7 Conclusions
7.1 Achievements
7.2 Perspectives
7.3 Conclusions
Bibliography
