Facial Expression Recognition Based on K-SVD Dictionary Learning and OMP Compression Perceptual Reconstruction
Xiaofang Jin, Ziyi Sha · 2021
Because of the non-contact and non-compulsory characteristics, emotional recognition has become a mainstream biometric technology, which is widely used in computer vision. As one of the branches of sentiment analysis, facial expression recognition is often affected by the attitude of the collector and the illumination of the environment. In the process of facial image collection, with the changing factors such as lighting and camera angle, the construction of a facial dictionary using samples will be affected to a certain extent. So that the test samples can not be linearly reconstructed by using the dictionary. This will lead to a decrease in the accuracy of recognition. To reduce the interference caused by misaligned faces and uneven gray distribution, a facial expression recognition method based on K-means Singular Value Decomposition (K-SVD) dictionary learning and Orthogonal Matching Pursuit (OMP) compressed perceptual reconstruction is proposed in this paper. Multi-task Convolutional Neural Networks (MTCNN) is used for face alignment and then use histogram equalization to preprocess. Finally, K-nearest Neighbor (K-NN) algorithm is used for facial expression recognition. After preprocessing, the new algorithm eliminates part of the nonlinear noise of the samples, and the recognition accuracy is improved from 20.8% to 67.8%, which improves the performance of facial expression recognition effectively.