Face Detection And Recognition Based On Two-Dimensional Image Representation

Face analysis is an important branch of biometric identification technology. It mainly deals with the automatic detection and recognition of human faces, etc. With the purpose of enabling the computer to replace human to achieve individual identity information quickly and accurately. This paper firstly summarized the application background, research contents and methods of face detection and recognition technology. Through the analysis of several typical algorithms in the past decade, a detection method of feature validation fusion was designed. A modified algorithm for face recognition based on local feature extraction was proposed.Firstly, we studied the face detection method based on Adaboost algorithm, carried on the introduction and summary of Adaboost algorithm in detail, and then the feature extraction operator used in face detection-- Haar-Like features were analyzed and discussed, and moreover the realization of weak classifier design, strong classifier structure and cascade detector design were given. The original main consideration of the traditional face detection algorithm based on the single Haar-Like features is the real-time property, rather than the ability to distinguish face and non-face. In view of this, we presented a method of MB-LBP feature verification fusion for face detection, and designed the realization steps of feature fusion. The method firstly obtains the candidate face regions identified by Haar-Like features, and then uses the face detector based on MB-LBP features for further screening, take the final output as the detected face region. Experimental results show that it is better in decreasing error detected faces.Secondly, the research of face recognition algorithm based on principal component analysis was carried out, a modified algorithm for face recognition based on local feature extraction was proposed, and an improved minimum distance classifier was adopted to realize the classification of samples. We first analyzed and discussed the mathematical principles and realization process of face recognition algorithm based on principal component analysis, pointed out the deficiencies as a statistical feature extraction based on holistic approach. On this basis, an improved version of two-dimensional principal component analysis(2DPCA) named bidirectional and modular fuzzy 2DPCA is proposed in this paper. This algorithm combined with the theory of modular matrix and bidirectional projection, the distribution information of overlapping samples were introduced into the scatter matrix in the form of weights allocation by fusing K-Nearest Neighbor algorithm, extracted lower dimensional local facial features from row and column directions with the space structure of image remain unchanged. Finally, experiments in ORL and AR face database proved that this method is better than other traditional methods in feature extraction, and has better adaptability under variations in illumination, time and facial expression.