Facial Expression Recognition using Low Level Histogram Features

The proposed framework recognizes facial expressions with the minimal histogram features. Emotional parts of the face are traced with the combination of face detection and facial landmark techniques; HOG integrated with a linear Support Vector Machine [SVM] classifier framework is adopted for face detection. The landmark points are aligned by using an ensemble of regression algorithm. The aligned facial landmark points are integrated by using a Delaunay triangulation. The mask is created by means of the convex hull points. The emotional parts of the face are isolated by the mask. For feature extraction, HOG and LDP descriptors are utilized to derive the distinct features of the detected face area. The dimensions of the fused feature vectors are reduced with the use of single-layer auto-encoder. The downsized features are recognized by using the soft-max layer. Facial expressions are recognized from still images and video frames. Images are collected from JAFFE, WSEFEP datasets, RAVEES audio-visual database and some of the images are collected from internet. The performance of the proposed model attains 95.7%.