Meng Ao

The HFB Face Database for Heterogeneous Face Biometrics research

A face database, composed of visual (VIS), near infrared (NIR) and three-dimensional (3D) face images, is collected. Called the HFB face database, it is released now to promote research and development of heterogeneous face biometrics (HFB). This release of version 1 contains a total of 992 images from 100 subjects; there are 4 VIS, 4 NIR, and 1 or 2 3D face images per subject. In this paper, we describe the apparatuses, environments and procedure of the data collection and present baseline performances of the standard PCA and LDA methods on the database.

Learning to fuse 3d+2d based face recognition at both feature and decision levels

2D intensity images and 3D shape models are both useful for face recognition, but in different ways. While algorithms have long been developed using 2D or 3D data, recently has seen work on combining both into multi-modal face biometrics to achieve higher performance. However, the fusion of the two modalities has mostly been at the decision level, based on scores obtained from independent 2D and 3D matchers. In this paper, we propose a systematic framework for fusing 2D and 3D face recognition at both feature and decision levels, by exploring synergies of the two modalities at these levels. The novelties are the following. First, we propose to use Local Binary Pattern (LBP) features to represent 3D faces and present a statistical learning procedure for feature selection and classifier learning. This leads to a matching engine for 3D face recognition. Second, we propose a statistical learning approach for fusing 2D and 3D based face recognition at both feature and decision levels. Experiments show that the fusion at both levels yields significantly better performance than fusion at the decision level.

A near-infrared image based face recognition system

In this paper, we present a near infrared (NIR) image based face recognition system. Firstly, we describe a design of NIR image capture device which minimizes influence of environmental lighting on face images. Both face and facial feature localization and face recognition are performed using local features with AdaBoost learning. An evaluation in real-world user scenario shows that the system achieves excellent accuracy, speed and usability

Near Infrared Face Based Biometric Key Binding

Biometric encryption is the basis for biometric template protection and information security. While existing methods are based on iris or fingerprint modality, face has so far been considered not reliable enough to meet the requirement for error correcting ability. In this paper, we present a novel biometric key binding method based on near infrared (NIR) face biometric. An enhanced BioHash algorithm is developed by imposing an NXOR mask onto the input to the subsequent error correcting code (ECC). This way, when combined with ECC and NIR face features, it enables reliable binding of face biometric features and the biometric key. Its ability for template protection and information cryptography is guarantied by the theory of encryption. The security level of NIR face recognition system is thereby improved. Experimental results show that the security benefit is gained with a sacrifice of 1-2% drop in the recognition performance.

Highly accurate and fast face recognition using near infrared images

In this paper, we present a highly accurate, realtime face recognition system for cooperative user applications. The novelties are: (1) a novel design of camera hardware, and (2) a learning based procedure for effective face and eye detection and recognition with the resulting imagery. The hardware minimizes environmental lighting and delivers face images with frontal lighting. This avoids many problems in subsequent face processing to a great extent. The face detection and recognition algorithms are based on a local feature representation. Statistical learning is applied to learn most effective features and classifiers for building face detection and recognition engines. The novel imaging system and the detection and recognition engines are integrated into a powerful face recognition system. Evaluated in real-world user scenario, a condition that is harder than a technology evaluation such as Face Recognition Vendor Tests (FRVT), the system has demonstrated excellent accuracy, speed and usability.

Face Recognition at a Distance: System Issues

Face recognition at a distance (FRAD) is one of the most challenging forms of face recognition applications. In this chapter, we analyze issues in FRAD system design, which are not addressed in near-distance face recognition, and present effective solutions for making FRAD systems for practical deployments. Evaluation of FRAD systems is discussed.

Bayesian Face Recognition Based on Markov Random Field Modeling

In this paper, a Bayesian method for face recognition is proposed based on Markov Random Fields (MRF) modeling. Constraints on image features as well as contextual relationships between them are explored and encoded into a cost function derived based on a statistical model of MRF. Gabor wavelet coefficients are used as the base features, and relationships between Gabor features at different pixel locations are used to provide higher order contextual constraints. The posterior probability of matching configuration is derived based on MRF modeling. Local search and discriminate analysis are used to evaluate local matches, and a contextual constraint is applied to evaluate mutual matches between local matches. The proposed MRF method provides a new perspective for modeling the face recognition problem. Experiments demonstrate promising results.

Nearest Feature Line: A Tangent Approximation

Nearest feature line (NFL) (S.Z. Li and J. Lu, 1999) is an efficient yet simple classification method for pattern recognition. This paper presents a theoretical analysis and interpretation of NFL from the perspective of manifold analysis, and explains the geometric nature of NFL based similarity measures. It is illustrated that NFL, nearest feature plane (NFP) and nearest feature space (NFS) are special cases of tangent approximation. Under the assumption of manifold, we introduce localized NFL (LNFL) and nearest feature spline (NFB) to further enhance classification ability and reduce computational complexity. The LNFL extends NFLs Euclidean distance to a manifold distance. And for NFB, feature lines are constructed along with a manifolds variation which is defined on a tangent bundle. The proposed methods are validated on a synthetic dataset and two standard face recognition databases (FRGC version 2 and FERET). Experimental results illustrate its efficiency and effectiveness.

Reducing Impact of Inaccurate User Feedback in Face Retrieval

A main problem in face retrieval is the semantic gap between low-level features and high-level semantic concepts. Relevance feedback (RF) may be used to incorporate to reduce the semantic gap. However, in the search for a specific target in a facial image database, a users assignment of RF instances may be mistaken. This would make the system prediction of the users target in a wrong way. Addressing this problem, we propose a new query point movement technique for target search by posing the problem of reducing the impact of inaccurate user feedback as an optimization problem. We develop a support vector machine based method to learn a decision boundary to identify ideal irrelevant images. Then we propose a rank function for finding target images, which would assign high scores to the images near the relevant images and punish those close to the decision boundary. Experiments are performed to show the stability and efficiency of the proposed algorithm.

Multi-Class Classification Based on Fisher Criteria with Weighted Distance

Linear discriminant analysis (LDA) is an efficient dimensionality reduction algorithm. In this paper we propose a new Fisher criteria with weighted distance (FCWWD) to find an optimal projection for multi-class classification tasks. We replace the classical linear function with a nonlinear weight function to describe the distances between samples in Fisher criteria. Whats more, we give a new algorithm based on this criteria along with a theoretical explanation that our algorithm benefits from an approximation of the ROC optimization. Experimental results demonstrate the efficiency of our method to improve the multi-class classification performance.