Research On Feature Extraction Algorithms For Face Recognition Under Complex Conditions

Human face recognition is an important research project in the province of image processing and pattern recognition. It has broad application prospects and commercial value in the field of identity verification, video surveillance, public security and digital entertainment.After decades of rapid development and the efforts of researchers, face recognition has achieved commercialize criterion in controlled conditions. While the accuracy of face recognition under uncontrolled conditions is seriously challenged by illumination variation, pose change, expression variation,object shelter, age change and so on. And the challenge of illumination variation is especially serious.Focusing on the problem of face recognition under illumination variation, we intensive researched face recognition method base on feature extraction and made a detailed analysis on existing classical method. And then we proposed two methods for face recognition base on Self quotient image feature and histograms of oriented gradients feature respectively.The primary innovations and works in this thesis can be summarized as follow:1.Based on Self Quotient Image, we import human face prior symmetry theory to recovering the face features in shadow or exposure regions caused by illumination variation to enhance the face recognition system rate and adaptability of complex environment. The paper designed several experiments to verify the effectiveness of the proposed method. First of all,it pre-processes the face images by using the illumination algorithm then extracts the advanced SQI face feature. After that, the dimensions of the advanced face features are reduced by Principal Component Analysis(PCA) and finally Support Vector Machine(SVM) is utilized in face classification and recognition.Experimental results based on Yale B standard face databases demonstrate that the proposed method is effective in eliminating the effect of the illuminate variation,especially under the circumstance that there are shadow or exposure regions in the face image.2.On the basis of histograms of oriented gradients feature, we proposed a novel approach for face recognition based on adaptively weighted HOG feature(AW-HOG) to solve the issues of low face recognition rate in complex illumination variation environments.Firstly, AW-HOG feature isavailable by fusing the weighting map and the traditional HOG feature of the sub-images divided from the original whole face images. And the weighting map is adaptively computed on account of the contribution of each sub-image. After that, the dimensions of AW-HOG features are reduced by PCA and the final classification features are generated. Finally,SVM is utilized in face classification and recognition using the final features.We designed several partitioning experiments and features comparison experiments to verification.Experimental results based on Yale B and AR standard face databases demonstrate that the proposed approach not only obviously enhances face recognition rate in complex environments but also has certain robustness to the influence of light and expression.3.According to the comprehensive study of face recognition techniques, we implemented a real-time face recognition entrance guard system based on QT The system consists of three modules: data registration,face recognition and system execution.The data registration module is mainly used to establish the guest database.The second module is used to identify the guest information which will influence the third module when controlling the status of entrance guard.When the system identifies success,it will open the entrance otherwise prohibit the entrance with alarm.