Representação esparsa e modelo de esparsidade conjunta no reconhecimento de faces

Abstract

This work proposes the usage of a Joint Sparsity Model with Matrix Completion (JSM-MC) for the composition of training set in the context of face recognition using the Sparse Representation-based Classifier (SRC). The proposed work aims to deal with face images in different illumination conditions and occlusions in the test and training set. For occlusions in the test set, an extended version of the algorithm is done to take into account occlusions in the optimization model. A pre-processing step is performed in the face images to reduce the effects of illumination change. A clustering of training images is done to reduce the processing time and a modification in the SRC algorithm is done to explore the sparsity of the sparse representation coefficients. The results are evaluated using a database with different illumination conditions. Artificial occlusions are inserted in the face images to investigate the behavior of the system in those conditions.