Research On Illumination Preprocessing Approaches In Face Recognition

Facial recognition has been one of the most successful applications in the field of imageanalysis. Although great breakthrough has been made with constant exploring in recent years,facial recognition faces greater challenge with the development of science and technology.Among many of the factors that influence the performance of the system, light has become abottleneck in the development of facial recognition. This thesis studies on the effect of light onthe facial recognition which will be illustrated as follows:Firstly, we analyzes the techniques of linear filtering and median filtering in image analysiswhich maybe effective for removing image noise but they can not been improved effectively,such as highlight. Hence, this paper has come up with a pretreatment of light based on imagedenoising which was built upon the square crossing center line of the image and also sought toaverage the gray scale of image toward outside. By this kinds of alternation method, we couldmake the pix gray value far away from the center O be approximate with the gray value of acertain element in the middle part, so as to achieve the goal of denoising.Secondly, employing the2DPCA feature extraction algorithm to experiment upon ORL andYale database. Compared with the recognition rate of90%without pretreatment, the recognitionrate of system comes to100%if the number of training and testing samples reaches5in thecase of Yale database. What is more, the effectiveness has been verified through integrating thealgorithm with the PCA, Fisherfaces、2DLDA feature extraction methods.Lastly, we had made a comparative analysis of other existing pretreatment methods, such ashistogram equalization, linear filtering, median filtering, Gamma gray correction andself-quotient image. By integrating with2DPCA feature extraction, we also made a comparativeverification between these common pretreatment methods upon the light and the pretreatmentbased on the image denoising.Our study has show that the pretreatment of image denoising has a better robustness interms of light effect and form transformation in the field of facial recognition and which alsoimproves the recognition rate accordingly.