基于最优相关滤波的光照容限二维子空间人脸识别

Face Recognition in Illumination Tolerant Two-dimensional Subspace Based on Optimum Correlation Filter

针对光照容限人脸识别问题,提出了基于最优相关滤波的光照容限二维子空间人脸识别方法。通过采用特定类2DPCA重构人脸图像,从中生成一对相关性过滤器。利用最优投影图像相关性过滤器将测试人脸图像投影到二维子空间,并利用重构相关性过滤器重构图像。根据预先设定的光照容限阈值进行人脸分类。在YaleB和PIE两大人脸数据库上对所提出方法的性能进行了评估,相比其他现存的相关性过滤器,所设计的滤波器更加优越。

For the illumination tolerant face recognition problem, illumination tolerant two-dimensional subspace method based on optimum correlation filter is proposed. A couple of correlation filters are generated by using 2D- PCA with class specified to reconstruct face images. The optimal projection image correlation filter is used to project testing images to two-dimensional subspace. Refactoring correlation filter is used to reconstruct images. Face classi- fication is finished by preset lighting tolerance threshold. The performance of proposed strategy is evaluated on YaleB and PIE face databases. Proposed technique has better performance comparing with other existing correlation filters.