Emotion recognition based on texture analysis of facial expression

The work being presented here portrays a novel approach for emotion recognition based on the texture analysis of facial images. The expression of the emotion on facial image manifests in variations in spatial arrangement and intensity of the pixels. The emotions can be recognized by taking note of these variations corresponding to the features being used for emotion detection in human interaction. In human interaction the most discriminating visual features of emotions are captured from lip, nose, eye, eye-brow and forehead segments of face. Hence in present approach the assimilated spatial and intensity variation of pixels are appreciated by the means of texture features extracted from the promising region of interest (ROI) on facial images. We have used MAZDA to extract texture features from the ROIs of the each class of emotions from JAFFE as well as In-House data set the most discriminating ten features were shortlisted using fisher discriminate analysis. These texture features were used for classifying the JAFFE and In-House data set using machine learning approach. A comparison is also being made between the proposed method and popular existing methods such as wavelet transform. The outcome of evaluation and comparison are included in original. The results are inclined to indicate that proposed method outperforms competing methods.