Research on facial emotion recognition based on eigenfaces
Facial emotions are usually categorized into seven categories:happiness,sadness,fear,disgust,anger,surprise,and normal.Due to uneven facial illumination,subtle emotion changes and other reasons,the accuracy of existing facial emotion recognition algorithms based on eigenfaces is low.Therefore,a new facial emotion recognition algorithm is established in this paper.Firstly,Viola-Jones algorithm was used to accurately detect and locate the facial region,and then Gauss filter was used to reduce the noise of the facial image,and Gamma correction was used to homogenize the illumination to obtain accurate and clear facial images.Secondly,the Haar-like features were used to accurately locate the center points of the left and right eyes,and the anthropometry method was introduced to locate and segment the facial emotion organs such as eyebrows,eyes and mouth,and then an eigenface is constructed in order to eliminate the information redundancy of non-emotional facial parts.Finally,LeNet-5 convolutional neural network was introduced to extract the deep digital features of eigenfaces for emotion recognition.Experimental results show that the proposed method can effectively improve the accuracy of facial emotion recognition.The accuracy on JAFFA public data set reaches up to 90.12％,which is better than that of geometric feature(53.75％)and full face feature(87.46％),and the performance is more stable.
万方数据
维普
