关键词: 人脸识别, 测地映射分析（GMA）, 核稀疏描述, 非线性空间

Abstract: Aiming at the problem that the face recognition system is vulnerable to the change of gesture, expression and occlusion in a non-controlled environment, a feature extraction method based on Geodesic Mapping Analysis（GMA） is proposed. Firstly, the similarity by calculating the geodetic distance between two pixels is measured. Then, the kernel sparse description model is built for the extracted GMA features, and the classification and recognition of the features are implemented in the nonlinear space. Experiments on the ORL and Yale-B face databases show that this method has a higher recognition rate in respect of free-form face images that deal with the change of severe occlusion, gesture and expression, which greatly improves the ability of face recognition system to deal with real complexed environment.

Key words: face recognition, Geodesic Mapping Analysis（GMA）, kernel sparse description, nonlinear space