Research Efforts on Recognition Methods

Face feature extraction and classifier designing are two essential parts for face recognition. For years, we have been working on methods including discriminant subspace analysis, Face Specific Subspace (FSS), Adaboosted Gabor Fisher Classifier (AGFC), Support Vectors based Kernel discriminant analysis (SV-KFDA), Local Gabor Binary Patterns (LGBP), Local Gabor Phase Pattern (LGPP).

  • Face Specific Subspace

Abstract: We present an individual appearance model based method, named Face-Specific Subspace (FSS), for recognizing human faces under variation in lighting, expression and viewpoint. This method is motivated, but essentially different from the traditional Eigenface. In Eigenface, each face image is represented as a point in a low dimensional face subspace shared by all faces; however, we experimentally show that one of the demerits of such a strategy is that the most discriminant features of a specific face are not accurately represented. Therefore, we propose to model each face by one individual face subspace, named Face-Specific Subspace. Distance from the face-specific subspace, that is, the reconstruction error, is then exploited as the similarity measurement for identification. Furthermore, to enable the proposed approach to solve the single example problem, a technique to derive multi-samples from one single example is further developed. Extensive experiments on several academic databases show that our method significantly outperforms Eigenface and template matching, which intensively indicate its robustness under variation in illumination, expression and viewpoint. See our IJIST2003 paper or contact Dr.Shiguang Shan.

  • Face Recognition Using Ada-Boosted Gabor Features

Abstract: Face representation based on Gabor features has attracted much attention and achieved great success in face recognition area for the advantages of the Gabor features. However, Gabor features currently adopted by most systems are redundant and too high dimensional. In this paper, we propose a face recognition method using AdaBoosted Gabor features, which are not only low dimensional but also discriminant. The main contribution of the paper lies in two points: (1) AdaBoost is successfully applied to face recognition by introducing the intra-face and extra-face difference space in the Gabor feature space; (2) An appropriate re-sampling scheme is adopted to deal with the imbalance between the amount of the positive samples and that of the negative samples. By using the proposed method, only hundreds of Gabor features are selected. Experiments on FERET database have shown that these hundreds of Gabor features are enough to achieve good performance comparable to that of methods using the complete set of Gabor features. See our FG2004 paper or contact Dr.Shiguang Shan.

  • Adaboosted Gabor Fisher Classifier (AGFC)

Abstract: Gabor features have been recognized as one of the most successful face representations, but it is too high dimensional for fast extraction and accurate classification. we have proposed the AdaBoost Gabor Fisher Classifier (AGFC) for robust face recognition, in which a chain AdaBoost learning method based on Bootstrap re-sampling is proposed and applied to face recognition with impressive recognition performance. In AGFC, AdaBoost is exploited to select optimally the most informative Gabor features (hereinafter as AdaGabor features). The selected low-dimensional AdaGabor features are then classified by Fisher discriminant analysis for final face identification. Our experiments on two large-scale face databases, FERET and CAS-PEAL (with 5789 images of 1040 subjects), have shown that the proposed method can effectively reduce the dimensionality of Gabor features and greatly increase the recognition accuracy. In addition, our experimental results show its robustness to variations in facial expression and accessories. See our AMFG2005 paper or contact Dr.Shiguang Shan.

  • Support Vectors based Kernel Discriminant Analysis

Abstract: We have proposed a novel nonlinear discriminant analysis method named by Kernerlized Maximum Average Margin Criterion (KMAMC), which has combined the idea of Support Vector Machine with the Kernel Fisher Discriminant Analysis (KFD). We also use a simple method to prove the relationship between both kernel methods. The difference of KMAMC from traditional KFD methods include (1) the within-class and between-class scatter matrices are computed based on the support vectors instead of all the samples; (2) multiple centers are exploited instead of the single center in computing the two scatter matrix; (3) the discriminant criteria is formulated as subtracting the trace of within-class scatter matrix from that of the between-class scatter matrix, therefore, the tedious singularity problem is avoided. These features have made KMAMC more practical for real-world applications. Our experiments on two face databases, the FERET and CAS-PEAL face database, have illustrated its excellent performance compared with some traditional methods such as Eigenface, Fisherface, and KFD. See our CVPR2005 paper or contact Mr.Baochang Zhang.

  • Local Gabor Binary Patterns (LGBP)

Abstract: For years, researchers in face recognition area have been representing and recognizing faces based on subspace discriminant analysis or statistical learning. Nevertheless, these approaches are always suffering from the generalizability problem. we have proposed a novel non-statistics based face representation approach, Local Gabor Binary Pattern Histogram Sequence (LGBPHS), in which training procedure is unnecessary to construct the face model, so that the generalizability problem is naturally avoided. In this approach, a face image is modeled as a “histogram sequence” by concatenating the histograms of all the local regions of all the local Gabor magnitude binary pattern maps. For recognition, histogram intersection is used to measure the similarity of different LGBPHSes and the nearest neighborhood is exploited for final classification. Additionally, we have further proposed to assign different weights for each histogram piece when measuring two LGBPHSes. Our experimental results on AR and FERET face database show the validity of the proposed approach especially for partially occluded face images, and more impressively, we have achieved the best result on FERET face database. See our ICCV2005 paper or contact Mr.Wenchao Zhang.

Abstract: Gabor features has been recognized as one of the most successful representation methods, such as Elastic Graph Matching, Gabor Fisher Classifier, and AdaBoost Gabor Fisher Classifier. One of the key issues in using Gabor features is how to efficiently reduce its high dimensionality. This paper proposes a multiple Fisher classifiers combination approach based on re-grouping Gabor features selected by using re-sampling and AdaBoost. At least two advantages can be observed with the proposed method: (1) more discriminative Gabor features are exploited in distributed Fisher classifiers by re-grouping the selected Gabor features; (2) combination of multiple Fisher classifiers improves the final performance of the classification compared with the traditional Fisher classifier. Our extensive experiments on two large face databases, FERET and CAS-PEAL, have impressively shown the effectiveness of the proposed method. See our BMVC2005 paper or contact Mr.Yu Su.