The Research On Color Constancy Theory And Application In Image Enhancement

Color is very important imformation in many research fields, for instance image enhancement, target recognition etc. Accurate color information is critical to further digital video image processing. Color constancy refers to a meaning that people have a stable psychological tendency that watching color of objects with change of the circumstances of light. Color constancy theory simulates the special function of human visual system to achieve image processing. This paper studies color constancy theory based on the calculation model. This article mainly shows as follows:1. The paper researched the basic theory of color constancy, including color image information theory, Retinex theory, gray world theory and Von Kries adaption. We analyzed some important algorithms of Retinex theory, for example random-walk Retinex, center around Retinex, white-patch Retinex. Experimental results are given and compared, we analyze and characteristics and limitations provide the new study inspiration.2. The linear weighted average method is adopted to integrate the results of different scale Retinex Algorithms, but it did not stress the merit of each scale in maintaining the details and the color fidelity effectively. So a novel Multi-Scale Retinex Algorithm proposed segments original image into four illuminance regions by involving the parameterized LIP model. Then each region is enhanced by Retinex Algorithm according to the illuminance difference. At last, based on proportionality factor, the enhanced subimages are fused. Experimental results show that the new algorithm is better than the traditional MSR method in both details and color fidelity.3. JPEG quantization tables for chroma and luminance obtain from Statistics and experiments of color images under natural light. But they don't have color correction function. So a novel JPEG compress framework introduced color constancy is proposed. It joins the Retinex theory to the JPEG compression process, estimating illumination through iterating local space average matrix got form DC coefficient. Last, according to illumination influence information and block edge direction information, the luminance quantization is modulated synthetically to correct color and compress images. Experimental results show that the new quantization table is better than the JPEG recommend quantization table in brightness and color fidelity with approximate compress rate.