Research Of Face Recognition Technology In Natural Environment

Under natural conditions of face recognition system, on the process face detection and face recognition should take into consideration the disturbance of environment, and the most typical is the interference of noise and illumination. Noise is largely due to wind, rain and snow weather formation noise in face sampling process and random noise in image transmission process; Light problem is mainly refers to collecting image process Angle and intensity of illumination are different, making face image too bright, too dark or the phenomenon of uneven illumination. Can properly solve the two factors of interference has an important influence to the recognition result. This paper completed the design of face recognition system, the main work content as follows:(1) Adopting the combination with skin tone and the characteristics method of face detection, adopting YCbCr color model which are high robustness to light, to reduce the illumination effect on face detection, through a certain scale of corrosion and expansion to reduce the noise influence on face feature. Finally realize more accurate of face detection.(2) Noise pretreatment. Select a mixture of salt and pepper noise and gaussian noise in the whole process of modeling, noise model are verified through the experiment to simulate the natural condition of noise has a certain effect, and the adaptive median filtering with wavelet transform method to eliminate noise, obtain good effect.(3) The pretreatment of illumination. Analysis of the conventional histogram equalization in extended image grey dynamic range, enhance the image contrast but at the same time it also merger grayscale, reduce the information entropy and fuzzy details. Based on above this paper propose a self-adaptive double platform histogram equalization to protect the information entropy algorithm. The algorithm based on double platform histogram equalization, for the purpose of maximizing to protect the information entropy. Using statistical difference iterative method iterative adaptive platform floor, and modified map function, to achieve a reasonable level gray mapping. Through the analysis of the objective evaluation parameters before and after processing, show that this algorithm overcomes the drawback of traditional histogram equalization and do well in processing the interference of light. Selection of logarithmic transformation model combining with Lee model algorithm processing of images.This method according with of human visual, based on this characteristic we select self-adaptive parameter to process the image of strong or weak light, in this way can make the details of face image is more abundant; By using the adaptive gamma transform better solve the problem encountered by the light, Through analyze of the objective evaluation parameters before and after processing to prove the effectiveness of the method.(4) Choose the improved LBP algorithm face recognition experiment, the first using LBP algorithm completed the image pixel level to feature level, the second with LTP characteristics of LBP algorithm to realize the feature extraction of image feature level, achieve higher human face recognition rate. Through the recognition rate of before and after image preprocessing validates the effectiveness of the pretreatment algorithm.