Facial expression recognition based on multi-radius local gradient binary pattern

The Facial Expression Recognition system recognizes the emotional state of a person by capturing the face, which is then filtered and matched with a pre-trained data-set. The automatic facial expression system impacts critical applications in many areas, such as consumer system for better human communication fields, customer satisfaction for product reviews in business sectors, dynamic texture recognition and aerial image analysis as well. This paper improves the traditional Local binary pattern and proposes a new feature vector named “Multi-radius Local Gradient Binary Pattern” (MLGBP). Initially, we extract the local\ninformation of facial region by calculating two Local Gradient Binary Pattern (LGBP) with adjacent radii and different kernels named LGBP_8;1 and LGBP_8;2 respectively. The proposed feature vector LGBP is more precise about the shadow and light effect of the face parts, which mainly decides the emotional states of a face. After that, we derive MLGBP by employing XOR operation on the LGBPs to find the correlate between them. This MLGBP feature vector can identify local gradient changes caused by light and noise by capturing information of various radii without an increment in feature dimension. We also studied the recognition rate of the proposed MLGBP feature vector with respect of several classifier and found that KNN provides the best result. Results of the CK+ database demonstrate the efficiency and effectiveness of the proposed system.