Research On Algorithm Of Facial Expression Recognition Based On Spatiotemporal Features

Face expression is an important channel for interpersonal communication.Video-based facial expression recognition has a wide range of application prospects in the areas of affective computing,human-computer interaction,public security,and psychotherapy.According to the differences in intensities and durations,facial expressions can be classified into regular expressions and micro-expressions.Extracting effective spatiotemporal features is crucial to improve the recognition performance.The main work of this dissertation is as follows:(1)We propose a feature learning method by combining the spatiotemporal features and geometric features learned by ISA network for facial expression recognition.Experimental results demonstrate that incorporating geometric features into spatiotemporal features can effectively improve the recognition accuracy of regular expressions and achieves better recognition performance compared to traditional hand-designed features.(2)A micro-expression recognition method based on spatiotemporal Gabor filters is proposed.The original videos are first preprocessed by Eulerian video magnification and frame difference,then the spatiotemporal features of the preprocessed videos of micro-expressions are extracted by Gabor filtering,the final recognition accuracy we obtained is better than LBP-TOP and HOOF.(3)We propose a new approach called STLGBP-TOP for micro-expression recognition.STLGBP-TOP extracts LBP-TOP on the responses volume of each spatiotemporal Gabor filter.Comparisons with LBP-TOP and spatiotemporal Gabor features demonstrate the effectiveness of the method.