Facial Expression Recognition by Independent Log-Gabor Component Analysis

Abstract

Research on facial expression recognition is critical for personalized human-computer interaction (HCI). Recent advances in localized, sparse and discriminative image feature descriptors have been proven to be promising in visual recognition, both statically and dynamically, making it quite useful for facial expression recognition. In this paper we show that the independent Log-Gabor feature (IGF), a localized and sparse representation of pattern of interest, can perform conveniently and satisfactorily for facial expression recognition task. In low-level feature extraction, Log-Gabor wavelet features are extracted, then ICA is applied to produce independent image bases that reduce the redundancy, emphasize edge information, while preserving orientation and scale selection property in the image data. In high-level classification, SVM classifies the propagated independent Log-Gabor features features as discriminative components. We demonstrate our algorithm on facial expression databases for recognition tasks, showing that the proposed method is accurate and more efficient than current approaches.