Face Recognition System Design

Face recognition system generally has two inevitable problems, one is the image brightness changes in image forming caused by uneven illumination, the second is the Gaussian noise produced in the image capture and transmission. Uneven illumination caused a dramatic change of image brightness, this brightness change have more impact on the face recognition than the differences of human facial feature. This paper completed the human face image recognition system, its main work is as follows:1, Processed the Gaussian noise of image using Kalman filter.2, Homomorphic filtering is an effective way to deal with image brightness change caused by uneven illumination, it can smooth image background, enhance the contrast and image detail. However, in actual use, due to the physical conditions of each image forming are different, so it is hard to find the best homomorphic filter parameters for each image. Fake Top-hat transformation can equally smooth image background, enhance image details shaded. During the experiment, we found the homomorphic filtering dependence on parameters was reduced, and the filtering effect also be enhanced in improving the homomorphic filter with a fake Top-hat transformation.3, Histogram equalization can map the gray-level histogram from a higher concentrated gray zone to the entire gray scale, solved the image brightness change caused by uneven illumination, though the intention is to enhance the information entropy of image, it still resulted enhancement of variance and contrast and expansion of gray dynamic range by reduced the image gray levels and information entropy. We designed an adaptive dual platform histogram equalization algorithm to protect entropy for lack of histogram equalization. The algorithm based on dual platform histogram equalization, first we calculated the lower limit platform for the purpose of information entropy maximization, then iterated the higher limit platform according to the nature of histogram equalization mapping function, finally modified cumulative distribution function to realize the reasonable gray-level map, it not only protected the image gray levels and information entropy, but also effectively improved image variance and contrast, extended the dynamic range of gray, overcame the disadvantages of histogram equalization.4, Correcting gray-level with adaptive Gamma value need to predefine illumination uniformity image, correct the comparable uneven illumination image using exhaustive method, and partly solve the image brightness change caused by uneven illumination. Effectiveness and efficiency of the algorithm exists a certain contradiction, this design used the magnitude approximation method to optimize the efficiency of the algorithm.5, Image recognition includes preprocessing, dimensionality reduction, projection, classification, identification. First used this design on the face image preprocessing, second used principal component analysis for dimensionality reduction and the projection, then used linear discriminant analysis for the optimal classification, finally identified the image class through the distance comparison. Choose Yale B face library and actual photograph of a face image to demonstrate the effectiveness of design in this paper.