Research On Algorithm Of Face Recognition Based On ID Photo

As a hot topic of the field of pattern recognition, face recognition is the most natural and direct identity authentication method, which has a good prospect of application in information security, access control and inspection and so on. But there are still some challenges of different aspects for face recognition, such as the impact on face recognition caused by posture changes, facial expression changes and light illumination changes. How to overcome the influence of these factors on the recognition rate has become the subject of the present study.The main content of this thesis is the identitfication face recognition, which has a certain constraint conditions, mainly to investigate the effect of expression changes, illumination changes and posture changes on face recognition. Specifically, this thesis is concerned with the research and experiment of face image preprocessing, feature extraction and classification.The main work of this thesis is as follows:1. In this thesis, we study the illumination compensation algorithm for face image with uneven illumination, and the effects of different illumination compensation algorithms are compared. As the experiments results show that, the illumination compensation algorithm based on single scale retinex can better enhance the edge of image, and it is advantageous to feature extraction and classification of human face image.2. Considering the characteristics and computational complexity of the feature extraction method, we selects the local binary pattern as the texture descriptor of the face image. Different characteristics of gray monotone invariance, rotation invariance, uniform pattern and rotation invariance plus uniform pattern are applied to extract LBP feature. The recognition rate and efficiency of these approaches are also compared by experiments. By analyzing data in experiments, the combination characteristic of rotation invariance and uniform pattern invariance has more accurate representation of the characteristics of the human face, which has the highest rate of face recognition considering the recognition rate and recognition efficiency.3. In this thesis, a classification model named GP-FC based on Gaussian Processes for binary classification is proposed. Furthermore, we combine the GP-FC model and the single scale Retinex illumination compensation algorithm. This classification model applied to face recognition with different expressions changes, light illumination changes and posture changes has better robustness. Compared with other face recognition methods, the recognition rate has improved 9-12 percent. Especially for face recognition with different posture changes, the recognition is as high as 98.77%.According to the application background of the identification face recognition, the classification model of this thesis has the ideal recognition effect, and it has good robustness to facial expression changes, illumination changes and posture changes.