Facial Expression Recognition by Transfer Learning for Small Datasets

Abstract

As a re-identification of facial attributes, facial expression recognition remains a challenging problem and the small datasets further exacerbate the task. Most previous works realize facial expression by fine-tuning the network pre-trained on a related domain. Therefore they have limitations inevitably. In this paper, we propose an optimal Feature Transfer Learning (FTL) algorithm to model the high-level neurons in a unified way. The proposed FTL structure is based on two models by correcting marginal distribution, matching the distribution between domains and optimizing the entire network connection by a parameter sharing method. Evaluation experiments based on three most public datasets of facial expression recognition: CK+, Oulu-CASIA and MMI, show that the proposed method is comparable to or better than most of the state-of-the-art approaches in both recognition accuracy and model size. Furthermore, we also demonstrate that our approach obtains more accurate results than other methods, such as directly fine-tuning a deeper network, training a shallower network from scratch.