Deep Face Emotion Recognition

Abstract

We extend the CNN based face emotion recognition to deal with the confusion of emotion recognition. We achieve state-of-the-art results on complex environment such as low or local light and blurry face details by using multiple input features fusion and mask loss which can focus on the valid local facial features, without any further refining and weighting multiple results module. Moreover, due to DenseNet construction of the model, our approach has much less parameters. Our method was tested on the Emotion Recognition in the Wild Challenge, Static Facial Expression Recognition sub-challenge (SFEW) and shown to provide a substantial 35.38% improvement over baseline results.