An Efficient Facial Expression Recognition System in Infrared Images

Most of the reported algorithms for facial expression recognition (FER) are based on visible expression databases. However, visible images are affected by illumination variations which can cause significant disparities in image appearance and texture. In this paper preliminary results of a new FER algorithm, using Gauss-Laguerre (GL) filter of circular harmonic wavelets to extract features for infrared images, are presented. By using GL filters with properly tuned-parameters, it is possible to generate a set of redundant wavelets that enable an accurate extraction of complex texture features from an infrared image. In addition, we utilize GL filters that are highly suitable for FER in visible images. The combination of infrared and visible FER using common feature extraction approach saves time and reduces the complexity in a multiple-sensors scenario. K-nearest neighbor is used for classification on OTCBVS and USTC-NVIE databases. The results show effective performance for using GL filters in FER for infrared images.