Research And Implementation Of Face Matching Algorithm Based On Semantic Feature

Face recognition has always been a hot topic in the field of pattern recognition and machine learning. In recent years, face recognition technology has made a great breakthrough, a lot of high efficiency and high accuracy of face recognition method is proposed. But these methods have not achieved the ideal recognition accuracy in practical application, and need further development and improvement. This paper focuses on the technology of face matching based on semantic feature, combined with face alignment method and semantic feature extraction method to analyse the shape feature and texture feature of face components,and use the semantic classification result of face components to recognise identity of face.In the face alignment, a face alignment algorithm based on 3D shape parameter regression is proposed, 3D face shape transform be divided into three-dimensional space transform and front face shape transform,in the two shape parameters regression framework, through iterative low dimensional 3D shape parameters to achieve face alignment. The method has higher alignment efficiency and alignment accuracy, and better alignment results are achieved in the 300-w complex face database. By contrast experiments, the superiority of our algorithm is verified.In the feature extraction, the facial features are expressed by six face components shape features and four face components texture features, including the eye shape, the eyebrow shape,the nose shape,the mouth shape,the contour shape, the face layout, the eye texture, the eyebrow texture, the nose texture and the mouth texture. In the shape feature analysis, user the 2D face shape to analyse the shape feature of face components, and build the shape model for each face component, and to define the corresponding semantic classification standard. Analysis of shape features using 2D face shape effectively reduce the influence of attitude factors on component shape. In the texture feature analysis, using 2D face shape to locate the position of the components and deduct the face components, using the Gabor+LBP feature extraction method to extract the texture features of each component, and through the kmeans algorithm for the semantic feature classification of each component.In face matching, the semantic feature encoding method is used to integrate the semantic features of the two faces, and the final matching result is calculated by the method of logistic regression.In this paper, we achieve the face matching algorithm based on the semantic features. We conducted a comparative experiment in the Web Face CASIA database. The experimental results show that our algorithm has a good matching effect, and achieves a good matching accuracy rate of 83.2% on the test database, which is better than the traditional Gabor, LBP and other face matching algorithms. At the same time, the algorithm has good stability and scalability, and can be directly used in the field of facial expression analysis, face search and so on.