稀疏表示人脸识别的关键问题分析

稀疏表示是一种高效的图像表示方法,且稀疏系数具有很好的稀疏性和可扩展性。基于稀疏表示的人脸识别能够提高识别率,增强鲁棒性。针对人脸识别在实际应用中遇到的问题,对稀疏表示人脸识别的方法、识别中遇到的关键问题及其解决办法进行综述。结果表明:稀疏表示人脸识别中,光照变化,可以通过增加不同光照的人脸图像训练样本解决;遮挡腐蚀,可以通过用加入误差字典来扩展过完备字典解决;姿势变化或未对准,可以通过对输入图像进行线性结构迭代变换解决;利用稀疏集中指数可以实现图像是否有效的判断。指出采用稀疏表示同时处理对准和连续遮挡的人脸图像识别,及识别准确性与实时性的提高是需进一步研究的方向。

A Survey of Face Recognition Based on Sparse Representation

Sparse representation is an efficient representation method for image, and the coefficient of sparse representation has good sparsity and scalability. It can make higher recognition rate and stronger robustness of face recognition. For problems of face recognition in practical application, the sparse representation method of face recognition, and the key problems in application and their solutions were summarized. The results show that： these problems can be solved in face recognition based on sparse representation method, such as illumination changes, by means of adding more train samples in different illumination; occlusion and corrosion, by means of using the extend over-complete dictionary with error dictionary; misalignment, by means of using linear structure iterative and using the sparsity concentration index to refuse invalid input image. To solve the problems of misalignment and continual occlusion in same time, and to improve the recognition rate and real-time processing are the directions of further research.