Face recognition is an important problem in pattern recognition and computer vision field. Due to various adverse factors such as low-resolution and heterogeneous modality variations, the performance of most existing face recognition algorithms is far behind satisfactory and hence limits its application and performance in real world. In addition, the increasingly serious situation of public security in the world needs more convenient identification and authentication technologies which can help or boost the traditional ones. For China, it is imperative to develop our own solutions with full intellectual property rights. Under this background, this thesis focuses on the main difficulties in face recognition research and attempts to provide a practical system design. Specifically, it proposes two algorithms which deal with low-resolution face recognition and heterogeneous face matching , respectively. And it also give a whole solution of real face recognition based on the new generation of embedded-system technologies, which is powerful, robust and configurable to be integrated to or cooperate with other working systems. The main contributions of this thesis include following issues: (1)To deal with low-resolution face recognition, we introduce supervised methods to subspace leaning and proposed a novel idea, e.g. simultaneous discriminant analysis(SDA). SDA learns two mappings from LR and HR images respectively to a common subspace where discrimination property is maximized. In SDA, the data gap between LR and HR is reduced by mapping into a common space; and the mapping is designed for preserving most discriminative information. (2)Coupled spectral regression (CSR) is an effective frame-work for heterogeneous face recognition. But in original CSR, the coupled projections are derived from the corresponding modality data respectively. The mutual information between different modalities are not sufficiently explored. So in this thesis, we propose to make up the projections by both(all) modality data, by which the discriminative information hidden in all samples are sufficiently explored. Moreover, the sample locality information is introduced and integrated into the proposed algorithm to improve its generalization ability. (3)Based on the dual-core processer TM320DM6446, the thesis propose a novel hardware design for practical face recognition system. This platform has powerful computation ability and plenty of resources and at the mean time i...
修改评论