The generalization ability of Convolutional neural networks (CNNs) for biometrics drops greatly due to the adverse effects of various occlusions. To this end, we propose a novel unified framework integrated the merits of both CNNs and graphical models to learn dynamic graph representations for occlusion problems in biometrics, called Dynamic Graph Representation (DGR). Convolutional features onto certain regions are re-crafted by a graph generator to establish the connections among the spatial parts of biometrics and build Feature Graphs based on these node representations. Each node of Feature Graphs corresponds to a specific part of the input image and the edges express the spatial relationships between parts. By analyzing the similarities between the nodes, the framework is able to adaptively remove the nodes representing the occluded parts. During dynamic graph matching, we propose a novel strategy to measure the distances of both nodes and adjacent matrixes. In this way, the proposed method is more convincing than CNNs-based methods because the dynamic graph method implies a more illustrative and reasonable inference of the biometrics decision. Experiments conducted on iris and face demonstrate the superiority of the proposed framework, which boosts the accuracy of occluded biometrics recognition by a large margin comparing with baseline methods.

/pdf/dynamic-graph-representation-for-occlusion-handling-in-319yfswv3d.pdf

Dynamic Graph Representation for Occlusion Handling in Biometrics

Efficient locality-constrained occlusion coding for face recognition

Abstract. We propose a local texture descriptor, local intensity area descriptor (LIAD), which is applied for human facial recognition in ideal and noisy conditions. Each facial image is divided into small regions from which LIAD histograms are extracted and concatenated into a single feature vector to represent the facial image. The recognition is performed using a nearest neighbor classifier with histogram intersection and chi-square statistics as dissimilarity measures. Experiments were conducted with LIAD using the ORL database of faces (Olivetti Research Laboratory, Cambridge), the Face94 face database, the Georgia Tech face database, and the FERET database. The results demonstrated the improvement in accuracy of our proposed descriptor compared to conventional descriptors [local binary pattern (LBP), uniform LBP, local ternary pattern, histogram of oriented gradients, and local directional pattern]. Moreover, the proposed descriptor was less sensitive to noise and had low histogram dimensionality. Thus, it is expected to be a powerful texture descriptor that can be used for various computer vision problems.

Local intensity area descriptor for facial recognition in ideal and noise conditions

Face recognition is a research hotspot in recent years. In order to improve recognition accuracy of face recognition, a feature selection method for face image based on Gabor feature and recursive feature elimination was proposed in this paper. Firstly, Gabor features were extracted from face image. Then, face image was divided into pieces and Gabor feature statistics of these pieces were linked in series to compose the original face image feature. Finally, recursive feature elimination based feature selection method was used to construct a low dimensional face image feature for face recognition. The proposed method was verified on ORL face database and the extended Yale face database B, and got high recognition accuracies. The experimental results show that this method can accomplish face recognition satisfactorily and is not sensitive to the inconsistency of details, such as facial expressions, poses and illuminations.

http://ieeexplore.ieee.org/iel7/6910079/6911426/06911497.pdf

Face Image Feature Selection Based on Gabor Feature and Recursive Feature Elimination

The generalization ability of Convolutional neural networks (CNNs) for biometrics drops greatly due to the adverse effects of various occlusions. To this end, we propose a novel unified framework integrated the merits of both CNNs and graphical models to learn dynamic graph representations for occlusion problems in biometrics, called Dynamic Graph Representation (DGR). Convolutional features onto certain regions are re-crafted by a graph generator to establish the connections among the spatial parts of biometrics and build Feature Graphs based on these node representations. Each node of Feature Graphs corresponds to a specific part of the input image and the edges express the spatial relationships between parts. By analyzing the similarities between the nodes, the framework is able to adaptively remove the nodes representing the occluded parts. During dynamic graph matching, we propose a novel strategy to measure the distances of both nodes and adjacent matrixes. In this way, the proposed method is more convincing than CNNs-based methods because the dynamic graph method implies a more illustrative and reasonable inference of the biometrics decision. Experiments conducted on iris and face demonstrate the superiority of the proposed framework, which boosts the accuracy of occluded biometrics recognition by a large margin comparing with baseline methods.The code is avaliable at this https URL\_Graph\_Representation

Dynamic Graph Representation for Partially Occluded Biometrics.

The human face is one of the most popular characteristic which can be used in the biometric security system to identify or verify a user. Face is an acceptable biometric modality because it can be captured from a distance, even without physical contact of the user being identified. Thus the identification or verification does not require cooperation of the user. Recognition systems based on human face are used for a wide variety of applications, due to these benefits. However, the crucial task is still to provide reliable recognition accuracy, but it is a challenging problem under real-world conditions. There have been proposed many methods, but only a few of them are being used in the real-world applications. Even the most recent face recognition algorithms are still facing problems when there is non-ideal imaging, varying illumination, occlusions in the scene or noise of used cameras. We solve these issues within The Next-Generation Hybrid Broadcast Broadband project (HBB-Next) [1]. In this project, we also deal with development of face recognition application, as part of multimodal interface, which will interact with HBB-TV user. In this paper we provide a comparative study of several conventional face recognition methods (PCA a.k.a. Eigenfaces, RBF) and novel kernel methods (KPCA, GDA and SVM) that are suitable to work properly under these conditions. We evaluate the influence of noise and partial occlusion on face recognition accuracy. We are focused on occlusions of eyes and eyebrows as these are the most significant features of a face. Face recognition rates achieved by machine learning methods with accuracy achieved by human perception only are compared. In addition we explore these methods for cases where only a few (up to 4) training samples is available.

Omelina Lubos

Papers

Face recognition under partial occlusion and noise