Adaptive joint block-weighted collaborative representation for facial expression recognition

Abstract

Facial expression recognition (FER) plays a signi cant role in human-computer interactions. Recently, regularized linear representation-based classi cation has achieved satisfying results in FER. Considering that different blocks in a sample should contribute differently to the representation and classi cation, we propose an adaptive joint block-weighted collaborative representation-based classi cation (JBW CRC) method to effectively exploit the similarity and distinctiveness of different blocks. In JBW CRC, samples are divided into different blocks and each block of the query sample is represented as a feature vector. Each feature vector is coded on its related block dictionary, which considers the similarity among the feature vectors. Additionally, the distinctiveness of different feature vectors is obtained by weighting its distance to other features, which addresses the distinctiveness in the different feature vectors. The proposed method is veri ed from the aspect of training samples, time complexity, and Gaussian noise variances on benchmark databases and the extensive experiments show that the proposed method is very competitive with some similar pattern classi cation methods.