Research Of Color Constancy Algorithm Based On Illumination Analysis

The scene illumination change has an obvious influence on imaging process of color image. The same object in the scene may present different image colors under different illumination conditions. This phenomenon has a very adverse impact on the study and application of some algorithms based on color features in the field of computer vision. Color constancy theory aims to solve the image color bias problem caused by illumination change. In this dissertation, the single light source color constancy theory and multiple light sources color constancy theory are in-depth studied systematically. The main contents are as follows:1. The research significance and practical application value of color constancy theory are introduced. The research status and the latest progress in the field of color constancy theory are teased out from the research of scholars at home and abroad. Furthermore, the theoretical basis and calculation model are introduced. The basic principles of several typical algorithms are expounded, and the advantages and disadvantages of these algorithms are analyzed.2. For the impact of the edge type and color channel correlation on the illumination estimation, a new single light source color constancy algorithm is proposed on the basis of image derivative framework. The proposed algorithm will firstly transform the derivative image into HSI color space, and calculate the saturation weight; Then derivative image is used to calculate the quasi-invariant, which is used to the classify the edges and calculate the edge type weight; Finally, saturation weight and edge weight are introduced into image derivative framework. Then the illumination estimate value can be calculated and the image can be corrected using von Kries diagonal matrix. The proposed algorithm has low complexity and comparatively obvious improved accuracy.3. For the widespread presence of multiple light sources in the real scenes, a mutilple light sources color constancy algorithm is proposed. The proposed algorithm firstly will divide the original image into blocks according to the grid-based partitioning strategy. Then select the appropriate single light source color constancy algorithm for each block according to the edge information richness itself. The selected algorithm is used to estimate the local illumination of the corresponding block. Finally, the obtained illumination estimates are combined to get several optimized illumination estimates for the light sources existing in the scene. The illumination value of each pixel in the image can be obtained by back-projection, and further the image can be corrected. The proposed algorithm could expand single light source algorithm to multiple light field. The algorithm accuracy is high and the practical value of color constancy algorithm is improved.4. This dissertation implements the experiments on several selected international general test image sets with the proposed algorithms. Experimental results show that compared with the previous classic color constancy algorithm, both algorithms perform better processing effect and the algorithm accuracy is much higher. Moreover, the proposed multiple light source algorithm can effectively solve the problem of color constancy for multiple light scene.