Human Emotion Distribution Learning from Face Images using CNN and LBC Features

摘要

Human emotion recognition from facial expressions depicted in images is an active area of research particularly for medical, security and human-computer interaction applications. Since there is no pure emotion, measuring the intensity of several possible emotions depicted in a facial expression image is a challenging task. Previous studies have dealt with this challenge by using label-distribution learning (LDL) and focusing on optimizing a conditional probability function that attempts to reduce the relative entropy of the predicted distribution with respect to the target distribution, which leads to a lack of generality of the model. In this work, we propose a deep learning framework for LDL that uses convolutional neural network (CNN) features to increase the generalization of the trained model. Our framework, which we call EDL-LBCNN, enhances the features extracted by CNNs by incorporating a local binary convolutional (LBC) layer to acquire texture information from the face images. We evaluate our EDL-LBCNN framework on the s-JAFFE dataset. Our experimental results show that the EDL-LBCNN framework can effectively deal with LDL for human emotion recognition and attain a stronger performance than that of state-of-the-art methods.