Implementation and experimental tests of illumination normalization using local variance in logarithmic domain
Date of Issue2016
School of Electrical and Electronic Engineering
After forty years research, face recognition technology has made significant progress, but there are still some unsatisfactory places. At present, many face recognition algorithms have been proposed, and achieved satisfying recognition results in a certain environment that the users are more compatible and the environmental conditions are more consistent. In Practice, illumination variation is a major problem in face recognition. In order to overcome the influence of illumination, two kinds of methods have been implemented in this report: using traditional image processing techniques to normalize the face image under varying lighting condition , such as histogram equalization (HE). logarithm transform (LT). Another method is illumination insensitive feature extraction method based on Retinex theory. The Retinex theory based on lambertian reflectance model is a classical illumination model consist of reflectance component and illumination component, are widely used in Gradient-Face, Weber-Face, illumination normalization using local variance in logarithm domain and multi-scale logarithm difference edgemaps. This report present the implementation and experimental test of illumination normalization using local variance in logarithmic domain and comparison with other approach. The result of experiments on both CMU-PIE and Yale B will indicate the quality of the illumination Normalization method in this project, which help researcher to improve illumination normalization on facial detection area in future.
Final Year Project (FYP)
Nanyang Technological University