Double OptconNet architecture based facial expression recognition in video processing

Abstract

ABSTRACT A Double optimization based convolution network model is introduced in the proposed video based facial expression recognition framework. The proposed model comprises U-shaped network, Residual-Network architecture, and Coot optimization. Before performing expression recognition, the input video is subjected to pre-processing, and face detection is performed over the extracted frames using the viola jones algorithm. The U-shaped network has the advantage of improving the processing speed of the convolution network, whereas the residual network can reduce the error that occurs during the frame encoding and gradient dissipation avoidance. Due to this merit, these two networks are combined and introduced in the proposed framework for facial expression recognition. The experimental evaluation is performed using a matrix laboratory tool over the three datasets: Affectiva-MIT Facial Expression Dataset, BAUM-1s and Real-world affective faces database. The comparative analysis shows that the proposed network has attained an efficient recognition rate than other existing network architecture.