Face Recognition Research Based On Improved Local Directional Pattern

Face recognition technology, a friendly biometric identification technology, has been applied in the field of identity verification. The research of face recognition algorithm is an important area in image processing, computer vision and pattern recognition. In recent years, face recognition has achieved satisfactory results under controlled conditions, through the efforts of many researchers. However, face recognition is seriously challenged by variations in illumination, pose, expression and occlusion, especially, variations in illumination.As an effective facial descriptor, local direction patterns(LDP) descriptor enhances the robustness of the local binary patterns(LBP) descriptor, but it has some inherent defects. Based on the analysis and comparison of LBP,LDP, we propose an improved local direction patterns algorithm. The main research work and innovative contributions are as follows:(1)A novel method of face recognition based on Nonsubsampled Shearlet Transform(NSST) and center-symmetric local directional pattern(CSLDP) was proposed. Firstly, a face image was decomposed with NSST and CSLDP operator was utilized to get CSLDP feature maps from sub-images. Then, CSLDP feature maps were divided into several blocks and the concatenated histogram calculated over each block was used as the face feature. Finally, nearest neighbor classifier was used to classify the faces. Experimental results on ORL, YALE CAS-PEAL-R1 face databases demonstrate that the proposed descriptor is simple and effective, and also robust to variations of posture, illumination and face expression.(2)An improved face descriptor ε – WLBD(ε Weber Local Binary Descriptor) based on texture features and gradient information extraction was proposed. Firstly, improved Local Binary Pattern was used for texture features extraction and improved Kirsch for edge features extraction. Then the histogram statistics were conducted respectively and concatenated into the general feature vector. The nearest neighbor classifier was used for face image classification and recognition. Compared with a variety of algorithms, experimental results on YALE face database and AR face database indicate that the proposed method is simple and effective, and robust to variations of face position, expression, occlusion and illumination, and also has better performance in face description for single sample.(3)A novel method of face recognition based on center-symmetric gradient magnitude and phase pattern(CSGMP) was proposed. Firstly, Gradient magnitude maps and phase maps of a face image was calculated. Secondly, a new operator named Center-symmetric Local Directional Pattern(CSLDP) was proposed to encode the gradient magnitude, and gradient phase was quantized into eight regions, then the proposed CSGMP was the combination of the binary codes of phase and CSLDP of magnitude. Finally, CSGMP feature maps were divided into several blocks, and the concatenated histogram calculated over each block was utilized as feature descriptor of face recognition. The recognition is performed by using the nearest neighbor classifier. Experimental results on YALE and AR face databases validate that the CSGMP algorithm is an outstanding method for single sample face recognition under different illumination condition, different facial expression condition and partial occlusion condition.