Research On Robust Face Recognition Algorithm Of Sparse Representation Based On Occlusion Detection And Recovery

Face recognition is a hot topic in the field of the computer vision because of its convenience as well as high efficiency. It is mainly used in the field including access control, video surveillance, social networking, photo management, criminal investigation,etc. In practice, however, the accuracy of face recognition algorithm is often impacted by the image occlusion. This paper will carry on some new research on the robust face recognition algorithm of sparse representation based on the basis of existing research results.First of all, we propose weighted fusion of multiple representations and changing dictionary for face recognition in order to eliminate the polluted, damaged pixels to reduce the adverse impacts on the face image classification and recognition. The new representation of image is generated by strengthening the moderate intensity pixels of the original samples, as well as weakening the influence of the other pixels at the same time.It is fused with the original image to form a new sample set. Finally, to create the changing dictionary which used for classification and recognition, occlusion detection of the test image is conducted to eliminate the occlusion of face image.Secondly, we raise a thesis that occlusion reconstruction and monogenic binary coding for face recognition due to the negative impact of the face image occlusion to classification and recognition. Down sampled sparse representation is used to the occlusion detection. And then we build face linear subspace and construct the over determined equation by Principal component analysis for occlusion reconstruction through the effective information. Finally, we encode the monogenic variation in each local region and monogenic feature in each pixel, and calculate histogram of the extracted local features for face recognition.Finally, face recognition algorithm based on local similarity statistics is proposed to impair the influence by occlusion to the accuracy of the face classification and recognition.Each face images in the database are divided into uniform blocks with no overlap by the same way and convert them to columns. We calculate the distances between eachsub-block of the test sample with the corresponding area of the training set, and then sum the smaller distances of the sub-blocks to get the final similarity for face classification and recognition.