Facial Expression Recognition Based on Semantic Patches

摘要

Recently, facial landmark detection has been well studied. Carrying on with these researches, we propose a semantic-patches-based framework in facial expression recognition (FER), which utilizes the information captured in facial vital senses. Firstly, the semantic patches centered on face landmarks are extracted. Then both Dense SIFT and corrected-Uniform Circle Local Binary Patterns (c-UCLBP) features are extracted and combined together. In addition, shape features denoting the relationships of facial components also extracted as a supplement. According to the proposed algorithm framework, we obtained good experimental results, and the accuracy on CK+ database and JAFFE database is 97% and 95% respectively.