A Preprocessing Method of Facial Expression Image under Different Illumination

Abstract

In this work, we propose an image processing method which combines the limited contrast adaptive histogram equalization (CLAHE) with Gamma transform to solve the illumination problem in facial expression recognition. We apply this algorithm to professional illumination datasets (Extended Yale B) and get better visual results, compared with using CLAHE and Gamma correction separately. Moreover, we use a convolution neural network (CNN) that pre-trained on FER2013 datasets to evaluate the effect of this method in facial expression recognition. We use this preprocessing algorithm to enhance the CK+ and Oulu expression datasets, and get accuracy of 89.24% and 70.24% respectively. Compared with the datasets that have not been pre-processed, it has provided an increase in classification accuracy of 7% on the Oulu datasets.