Review and Comparison of Random Spray Retinex and of its variants STRESS and QBRIX

In this paper, we review and compare three spatial color algorithms of the Milano Retinex family: Random Spray Retinex (RSR) and its subsequent variants STRESS and QBRIX. These algorithms process the colors of any input image in line with the principles of the Retinex theory, introduced about 50 years ago by Land and McCann to explain how humans see colors. According to this theory, RSR, STRESS and QBRIX re-scale independently the color intensities of each pixel by a quantity, named local reference white, which depends on the spatial arrangement of the colors in the pixel surround. The output is a new color enhanced image that generally has a higher brightness and more visible details than the input one. RSR, STRESS and QBRIX adopt different models of spatial arrangement and implement different equations for the computation of the local reference white, so that they produce different enhanced images. We propose a comparative analysis of their performance based on numerical measures of the image brightness, details and dynamic range. In order to enable result repeatability and further comparisons, we use a set of images publicly available on the net.