Sparse Representation for Face Recognition

Face recognition has been one of the most extensively studied problems in the area of pattern recognition and computer vision. In a typical face recognition system, a gallery of training face images of multiple individuals under different illumination conditions or with different facial expressions are provided. The goal is to identify a given test face image as one of the individuals in the gallery set. Generally, the dimension of face images is very high, thus leading to high computational cost. Robust face recognition via sparse representation, provides a new solution to work with high-dimensional face images through exploiting their sparse structures using the theory of compressed sensing. This technique tries to use the fewest possible training images from the gallery set to interpret a given test face image. Treating the gallery training set as a large dictionary, sparse presentation-based classification aims to approximate a given test face image by seeking a sparse linear combination in terms of the dictionary and images of individual pixels. The most compact representation using training images from a certain individual can well encode the identity of the test image. The sparse representation framework achieves a striking recognition performance even with severe occlusion and corruption presented on the face images.