自动人脸识别关键算法分析与实现-analysis and implementation of key algorithms for automatic face recognition.docxVIP

华中科技大学硕士学位论文 华 中 科 技 大 学 硕 士 学 位 论 文 II II Abstract The current face recognition algorithms has been relatively mature. Through most kinds of recognition algorithms has been relatively effective on accuracy, they have low time efficiency. This article tries to improve the time efficiency of the algorithms on the basis of not reducing the accuracy of the existing algorithms . The main research contents and achievements of this dissertation are as follows: The dissertation researches on the basic theory of image processing, including canny edge detection, hough transform, image rotation transformation, which are related with face recognition. At the part of eye location by the method of face image hough transform, semi-circle detection is used by this algorithm other than circle detection, and good results have been achieved. The dissertation researches the cascade classifier based on adaboost algorithm structure, using face-haar-features to train the cascade classifier and to detect face regions. For the non-face-region which has similar face-haar-features, this algorithm is prone to classify it to face region, this dissertation proposes a method to eliminate the non-face-region by skin color detection based on YCbCr pattern, and puts forward rapid skin detection to save run time, to ensure the system in real-time. In the process of extracting face image features, the Gabor wavelet features of eight direction and five dimensions have been extraction. For the problem of so many features, this dissertation has adopt down-sampling method and PCA algorithm. Experimental results show that this method has eliminated great many uncorrelated elements and greatly simplified the structure of recognition system. Artificial neural network based on back propagation algorithm(BP-ANN) has been structured, by the way, sigmoid unit as perception has been adopted, delta rule has been used to training network, and the method of adding momentum factor has been employed to improves the l