A New Facial Expression Recognition Algorithm Based on DWT Feature Extraction and Selection

Abstract

In this paper, we propose an efficient framework to improve accuracy and computational cost of a Facial Expression Recognition (FER) system. This framework is carried out in three stages. In the initial one, corresponding to feature extraction, three descriptors, derived from Discrete Wavelet Transform (DWT), are introduced to extract distinct feature types. In the second stage, focused on feature selection, a Wrapper approach is adopted to carefully select the most relevant features from the previously extracted pool. Following feature selection, the Support Vector Machine (SVM) classifier is employed, in the final stage, to determine an individual's affective state. The experiments were conducted in person-independent mode using both the Japanese Female Facial Expression (JAFFE) and extended Cohn-Kanade (CK+) databases which included the following emotions: anger, disgust, contempt, fear, happy, sad, surprise, and neutral. The obtained results demonstrated the effectiveness of the proposed framework in increasing recognition rate and decreasing response time compared to other state-of-the-art methods. A comparative study between our proposed framework and that based on the Local Binary Patterns (LBP) method demonstrated that our framework outperforms the latter for most emotions. In fact, our proposed framework converges rapidly and achieves good performance, thus allowing us to develop a real-time Facial Expression Recognition (FER) system in person-independent mode. Average recognition rates of 89.66% and 87.76% were obtained using our method with the JAFFE database and the CK+ database, respectively.