Facial beauty prediction combined with local binary pattern and convolutional neural network

Convolution Neural Network (CNN) for facial beauty prediction, can learn the deep feature expression, but can extract the global feature and neglect the local information of the face. Therefore, it has poor generalization ability. In this paper, a facial beauty prediction algorithm combined Local binary pattern (LBP) and CNN is presented. Fistly, data augmentation technology is utilized to expand the scale of the database. Secondly, the LBP texture image is channel-fused with the original grayscale image, and then the linear combination of channel feature maps is implemented by a 1×1 convolution operation. Thus the cross-channel information fusion of the network is realized, so as to improve the accuracy of facial beauty prediction. Experimental results based on the Large Scale Asian Female Beauty Database (LSAFBD) show that the algorithm presented has good prediction ability in the classification and regression, which is superior to the other models for facial beauty prediction, and demonstrate that adding texture images to CNN can effectively improve the accuracy of facial beauty prediction.