Facial Sub-regions for Automatic Emotion Recognition using Local Binary Patterns

Facial expressions are considered as a relevant source of information in order to automatically detect and recognize the human emotional state changes. Over the last decades, numerous methods have been proposed. One of the most popular and widely used descriptors remains the Local Binary Patterns. However, the performance of the existing methods in terms of accuracy varies and needs to be improved. In this paper, we introduce an approach that exploits specific racial sub-regions in order to generate a spatial representation using the Local Binary Patterns technique. Moreover, two different dimension reduction techniques (namely Principal Component Analysis and Independent Component Analysis) are used to improve the generated representation. The recognition of the six basic emotions is achieved using a multi-class Support Vector Machine classifier. The obtained results after validation with three benchmark datasets attests to the efficiency of the proposed approach since it yields 96.63%, 94.25% and 86.19% with the JAFFE, RaFD and KDEF datasets, respectively.