Research On The Technology Of Face Recognition And The Design Of Software System

Although, automatic face recognition technology has been studied for many years, it is still a hot spot in computer vision. The paper focuses on some key technologies in the field of face recognition including the face detection, feature extraction, design of classifier and the realization of software platform.In terms of face detection, the theoretical knowledge of AdaBoost algorithm is analyzed. Based on these basics, the paper uses the OpenCV source platform to detect the face from an image and then normalizes it to correct the distortion introduced by the rotation, the size and the gray. Experimental results on FERET face databases show that the identification accuracy can be improved by the image standardization.In feature extraction, the theory of PCA, LBP, LGBPHS and Weber Local Descriptor(WLD) are mainly suggested. The advantages and disadvantages of them are also analyzed. Among them, the performance of WLD is obviously superior to other algorithms. But WLD feature is the histogram feature which is a kind of global descriptor. It focuses on describing the overall properties of the image and is easy to lose the detailed information. We improve the performance of WLD by introducing local spatial information. Besides, the method to calculate the gradient orientation is improved to do better for extracting the direction information and suppressing noise. The Support Vector Machine(SVM) is used to verify the performance of improved WLD. Experimental results on ORL and YALE face databases show that the improved WLD feature can enhance the face recognition accuracy obviously. At the same time, the paper also presents the theory of nearest neighbor classifier, and the performance of them is verified by the LBP. The corresponding experimental results show that the SVM is superior to the nearest neighbor classifier.Finally, the face recognition software system is built based on the requirement of the experiment. The processes which include face images real-time collection, face detection, face images training and face recognition are finished under the real scene. Experimental results demonstrated the feasibility and effectiveness of related works. It achieves the expected effect.