Secondary abstract: |
Photography, printing and computer vision are the areas where color constancy of images is essential. The human beings have learned to eliminate the effect that illumination has on the color appearance of objects, which is also the goal of color constancy algorithms. In this diploma thesis, we describe modern algorithms of the two-level process of color constancy, which consists of illumination detection and image transformation. We propose a new method for image transformation based on the use of color subspaces. The objective comparison of the illumination detection algorithms shows that the Bayesian color constancy algorithm proves to be the most effective. In the case of image with known illumination, the suggested algorithm for image transformation with the use of color subspaces proves to be the most effective. Visual analysis of images transformed with the method of using color subspaces, reveals two deficiencies of the method -- noise sensitivity and sharp color refractions. In the case of image with unknown illumination, all the methods for image transformation prove to be equally effective. The correct illumination detection has a larger effect on the transformed image as the used method of transformation. In the case of incorrect illumination detection, the visual appearance of images transformed with the color subspace method proves to look more natural. |