Study Of Face Recognition Feature Extraction Algorithm

Face recognition is a hot research field in biometrics identification technology, and there are many ways to realize it. How to implement it accurately and efficiently is one of the major difficulties in face recognition. Generally, a complete face recognition system consists of four main areas: the acquisition of face images, face detection, facial feature extraction and feature matching. In these areas, how to detect face fast and accurately, and how to extract facial feature accurately will affect the performance of the system directly. In this paper, we select the existing classic face detection and facial feature extraction algorithm to analyze, research and implementation. To solve the problems found during the experiment, we improved the classical algorithm combining with knowledge of digital image processing and pattern recognition. The performance of these algorithms has been improved to a certain degree.The major research work is as follows:1. Described a face detection method based on the AdaBoost algorithm, which has the most practical value at present. Researched on how to extract Haar-like features, how to design classifiers and how to solve the problem on multi-scale transform. Meanwhile, in order to further reduce the false detection rate, we took skin color calibration on the results of AdaBoost algorithm to remove the false face areas effectively.2. Studied and implemented the facial feature extraction algorithm base on active shape model (ASM). Analyzed the model building and matching search algorithm of the classical ASM in detail. To solve the difficulties encountered in the traditional method with expression in the face, we proposed an improved ASM algorithm: divide face region by change-related degree and model and search independently. The improved algorithm can achieve a more accurate feature point location with the existence of human facial expressions or in the case of talking. In addition, we also tried to replace principal component analysis with kernel principal component analysis to improve the location accuracy under the situation of face with small-angle deflection.3. Researched and implemented facial feature extraction algorithm based on Active Appearance Model (AAM). Analyzed the modeling and matching search algorithm of the classical AAM in detail. Studied how to using Inverse Compositional Algorithm in the matching searches process. To reduce the interference taking by global illumination, we took the texture normalized operation into the inverse compositional algorithm integrated, and achieved an improvement on the accuracy of the model instance.4. Achieved the face detection and facial feature extraction algorithm proposed in this paper with OpenCV, and built a complete face recognition demo using MFC.