Research On Image Feature Extraction And Matching Technology In Face Recognition

With the rapid development of information technology, artificial intelligence, pattern recognition, computer vision and other new technologies,face recognition has been widely used in public safety, information security, finance and other fields. It has become one of the hot topics of research in image processing fields in recent years. For a face recognition system, the purpose of image matching is to unify these face images from different sensors, different times, different visual scenes into a framework, in order to facilitate subsequent feature extraction and recognition. Therefore, feature extraction is the premise and basis of face recognition technology, it is very critical to study some fast and efficient methods for human facial feature extraction.The main aim of this work is to propose feature extraction and image matching techniques. With face recognition as application background, different feature extraction methods will be explored deeply. The main contents and results are given as follows:(1) The aim of image matching is to implement two or more images to be aligned in the same space, which is used to determine the existence transformation relationship among them, and these images are from different time, different sensor and different scene. The basic definition and process of image matching are described, and the image matching method based on projection entropy is given in detail.(2) A new and robust face recognition method is proposed to overcome the flaw with low recognition rate to be used singular value vector-based recognition method. The class estimated basis space method is used to extract facial singular value, and then a method fuses multi-scale global and local features in order to reduce the effects of illumination, expression, noise, pose and other factors.The rough set reduction algorithm is used to select feature, and ultimately selected features are as the input into the SVM classifier. Experimental results show that the improved method has the validity, and the method fused multi-scale integration of the global and local feature is a more efficient feature extraction method. (3) To illumination challenge in face recognition, the illumination processing method based on total variation model is given in detail. On the basis of total variation model, the L1optimization combined with Bregman Iterative Algorithm is used to solve the reflection coefficient. In order to improve the processing speed of the model, the multi-resolution method is applied to find optimal solution from a rough value to the best solution. Finally, modular PCA method is used to extract the local features to describe difference and similarity among face images more effectively. Experimental results show the effectiveness of the method, and the improved method has obtained better recognition rate.(4) Intensity values used in original ASM can’t provide enough information for model searching, which is also sensitive to lighting conditions and so on. So we use a measure which indicates the orientation of structure at each pixel instead of intensity value to represent image texture. In addition, a new method is adopted for building local profile model, which makes full use of texture information around the landmarks. Experimental results show that the improved ASM can locate face features more accurately than original ASM, and face recognition rate is much larger based on the improved ASM feature extraction method than that of original ASM.(5) In original ASM model, the Principal Component Analysis (PCA) approach is used to extract shape eigenvectors of the training data. The traditional PCA method can’t real-time to update the covariance matrix and mean texture, which is sensitive to pose, illumination and expression variations in images. An improved method is propose to overcome the flaw, the feature space is constantly updated by using the incremental principal component analysis (IPCA), which it can describe the similarity or difference among training image sets. The experimental results show that the improved method can effectively improve the matching accuracy and it can also improve the face recognition rate.In this thesis, some improved feature extraction methods are proposed by deeply researching about feature extraction algorithm and image matching technology for face recognition. To overcome these flaws of singular value vector-based recognition method, a new and robust face recognition method based on fusion global and local feature is proposed. To illumination challenge in face recognition, a novel and robust feature extraction method based on total variation model is constructed. After analyzing the basis of statistical model, two improved ASM can locate facial feature more accurately. By the way of combining theory with experimental results, it is a useful exploration process for the deep study of relative fields of face recognition and feature extraction.