Abstract: Nonuniformity of luminance in images due to irregular lighting etc. could cause difficulties in various kinds of image processing in face detection. A normalization method was presented for recognizing human faces under variation in lighting, which was called local autocorrelation (LAC) method. LAC method was applied to human face detection based on Adaboost algorithm. The classification result of CMU PIE database for original and LAC images were compared with the LAC method. The physical properties of the LAC were analyzed, and the LAC robustness of linear changes in illumination was verified theoretically. Experimental results showed the number of weak classifiers could be reduced to a great extent, while preserving equal detection capability. The effectiveness of elimination of nonuniform illumination variation in images was verified in face detection experiment.

Key words: local autocorrelation, illumination variation, image preprocessing, face detection, Adaboost algorithm