Proposed approach of detecting facial emotion using neural network and representational of HOG features

The subject of emotion detection from digital image has gained exceptional importance in recent years due to the expansion of visual applications in variowefields of life. With respect to the emotion of human face, the matter becomes more complex according to its variety. At the same time, this matter becomes easier if the computerized technique is used to learn most known emotions and then detect it in the final imaging system. In this paper, a new approach for detecting emotion of human face has been proposed using artificial neural network (ANN). This network is feed by a set of histogram of gradient (HOG) features, as a representative reference to describe the entire emotion. The determining of HOG features is limited to specific region of the face within the digital image. This region is designed to take T shape which covers main parts of human face (eye, noise, mouth, and eyebrow) that are changed with emotion type. The proposed approach is evaluated by standard emotion dataset (JAFFE) in both phases of ANN (training and testing). The simulation results view significant percentage of accuracy in comparison with the existing technique of emotion detection.