Reduction of illumination effect by using adaptive histogram equalization in face recognition

Face recognition is a widely studied field in recent years due to its important role in many applications, such as law, criminal identification, credit card verification, security system and intelligent surveillance. Although present methods have good performances under certain conditions, there are several problems such as: different poses, different illumination conditions and changing expressions, that affect the performance of the face recognition methods. The illumination variation is one of the most difficult problem facing researchers. In our thesis, we investigate different illumination normalization methods based on histogram equalization applied globally, locally and adaptively for face recognition. In order to increase the performance, we apply different filters on the histogram equalized images, then we compare the following methods: Histogram Equalization (HE), Histogram Equalization with High Boost filter (HE+HB), Histogram Equalization with Laplacian filter (HE+Lap), Local Histogram Equalization (LHE), Local Histogram Equalization with Median filter (LHE+Med), Local Histogram Equalization with Gaussian filter (LHE+Gaus) and Adaptive Histogram Equalization. These methods can be used to eliminate the effect of uneven lighting condition effectively and efficiently, and improve the recognition performance. In the experiments which are evaluated and compared on the Yale face database B, local histogram equalization is adapted with different window sizes and different parameters. The experimental results show that the face recognition rate is improved from 46.49% (for images with no processing) to 99.55% (for LHE[7 7]+Gaussian 3x3 with standard deviation=1 method) with an error rate 0.45%. The key point is that this success is obtained with a simple method.