Classifier feature extraction techniques for face recognition system under variable illumination conditions

Abstract

Recognition of object under uncontrolled illumination environment is imprecise and vague. A simple image preprocessing chain is taken for precept. Local binary pattern (LBP) is capable of reducing noise levels in background regions. Local ternary patterns (LTP) fragmenting less under noise in uniform regions. Gabor filter acts as a resounding filtering source for local spatial frequencies. Phase congruency is to extract the image in phase as well as in magnitude levels. The result is obtained by the KLDA based classifiers with combination of LBP and Gabor features. The above explained features are obtained from both the input and the data base image. In that the LBP and Gabor features are fused and the distance is calculated. If both the input and database images are same, the face is recognized; otherwise the face is not recognized. The simulation results exemplify the proposed technique for image with different lighting, expressions and structural defects. © 2011 Springer-Verlag.