A holistic approach towards detection, tracking, and recognition of face

To detect and track the face in a complex environment is still challenging due to variations in speed, illumination, occlusion, rotation, scaling, pose, and imposters. Therefore, a computationally efficient model for face and facial parts detection/tracking is proposed in this work. A deep convolutional neural network (DCNN) based model is designed to detect the face and track it in the successive frames using a point-tracker. The DCNN is developed by re-designing the AlexNet network and tune the weights corresponding to the new classes. A shallow architecture is designed to recognize the face by utilizing hand-craft features and proposed a new distance feature. This model achieves a relative improvement of ~8% over the baseline models. To evaluate the performance of the proposed model, five image databases are used, i.e., the ORL face database, Yale face database, JAFFE database, CK, and the LNMIIT database. The proposed models achieved significant detection and recognition accuracy at a low computation time.