Facial expression recognition using singular values and wavelet‐based LGC‐HD operator

Abstract

Abstract Facial expression recognition (FER) is a significant research area in the human–computer interaction. The performance of FER systems depends on an efficient feature extraction method. This research work proposes a robust method for feature extraction using local gradient coding based on horizontal and diagonal (LGC‐HD) operator, wavelet transform and singular values. The proposed framework consists of three steps in feature extraction. First, the wavelet transform is applied on facial images and features are extracted by applying the LGC‐HD operator on low–low bands at different levels of decomposition. Second, the singular values of the facial image are calculated by applying singular value decomposition and these are used as features. Third, the features from the previous two steps are concatenated to obtain the final feature for FER systems. The proposed method combines the two advantages: (1) the integration of wavelet transform and LGC‐HD operator provides discriminating and robust features. (2) The singular values are not sensitive to grey scale changes caused by the noise. After obtaining the features, the support vector machine is used for expression recognition. Extensive experimentation on Japanese Female Facial Expression database, Cohn–Kanade, FEI face and facial expression research group databases achieves good recognition rates of 84.2%, 92.3%, 98.2% and 97.9%, respectively.

Keywords

• emotion recognition
• face recognition
• feature extraction
• singular value decomposition
• support vector machines
• wavelet transforms