A New Compressive Sensing Method for Face Recognition

We propose a compressive sensing based method to recognize face with pose variations. The face recognition framework includes two issues: feature extraction and classification. For feature extraction, we present a random measurement matrix to compress an image from high dimensional space to a low dimensional space. The compressive feature has a powerful discrimination because it can preserve most salient information of the image. Meanwhile, the random measurement matrix requires only a uniform random generator. Consequently, the computational complexity is very low. For face classification, we adopt a gradient projection approach considering the Barzilai-Borwein steps and adaptive nonmonotone line searching method. The approach can not only guarantee global convergence but also keep the gradient projection's performance. Finally, our method is conducted on the ORL face database. The results illustrate that our method can handle the variations pose and perform well in term of computing time and recognition rate.