An expression detection technique based on multi-input convolutional neural network for incomplete face images

摘要

An expression detection technique based on feature fusion multi-input convolutional neural network is proposed. In view of the negative effect of occlusion objects in occlusion face image on expression recognition task, the multi-input convolutional neural network is proposed to use the multi-input property, so that the multi-classifier coupling network can learn more complex prediction model. The local feature level fusion method was used to extract the features from the image, and the local micro-features of the region of interest were taken as the multi-branch input of the multi-input neural network, so as to reduce the influence of the contribution rate of the missing part of the incomplete image and improve the robustness and accuracy of the expression detection.