Appearance improvement of color image by adaptive linear retinex model

The electronic camera can't catch the details in the heavy change of highlight and shadow, while human vision can see a very wide range of luminance scenes with keeping the color constancy. This paper presents a method to improve the image appearance based on Adaptive Scale-Gain MSR (Multi-Scale Retinex) model and discusses the design of optimum parameters based on visual appearance experiments. Conventional Retinex model needs for careful adjustment in parameters and hasn't any clear target to be reproduced. In this paper we get a target image by adjusting the camera image on screen to be visually close to a test scene set up in our laboratory. The kernel sizes and weights for MSR are optimized to match the target image in minimum color difference.