Image quality measurement-based comparative analysis of illumination compensation methods for face image normalization

Illumination invariance is one of the most challenging features to obtain in the facial recognition system. Normalization of lightening is a way to solve this problem. Low light is enhanced for pre-processing of the input image to improve image quality. Various histogram-based image enhancement methods such as histogram stretching (HS), local histogram equalization (LHE), adaptive histogram equalization (AHE), histogram equalization (HE), and global histogram equalization (GHE) are applied based on image quality measurement methods, i.e., normalized absolute error, correlation, peak signal-to-noise ratio (PSNR), average difference mean-squared error (MSE), and structural content. Non-reference image quality assessment test and subjective quality test are accomplished. It is observed from the statistical results that the GHE gives better performance in contrast to improvement. GHE image will undergo a point-based transformation, i.e., logarithmic transform. This will be followed by an equalized image with discrete cosine transform (DCT) and a relevant DCT image factor modifying low and high-frequency coefficients. Since the objects taken in this paper are face images with no background, this method will give excellent results. Statistical significance test shows that the differences/improvements are significant. Moreover, the dynamic range of face images is improved using logarithmic transformations. It can be seen from the results that for the face images, with very high illumination dissimilarities, the proposed method improves the performance significantly.