Research On Methods Of Illumination Color Compensation

Color appearance of objects is dependent on the illuminant and surface characteristic of objects. The same object shows different colors under different illumination. Illumination color compensation technology, which studies how to describe intrinsic color of objects under different illumination, has become the research hotspot in the computer vision fields.This thesis mainly studies illumination color compensation and objective evaluation methods, and focuses on illumination color compensation technology, which transforms a source image while being taken under an unknown illuminant to a target image obtained by the same camera under a standard illuminant (such as D65), to eliminate the effects of illumination.A new color difference calculation method based on RGB color space was designed. The experiment results show the effectiveness of the proposed color difference formula.Two different illumination color compensation algorithms were proposed for different applications. The illumination color compensation algorithm based on color space transformation, estimates the illumination color and then transforms the RGB cube to a color space XLYLZL. The transformation matrix which the original image is converted to the target image is obtained. The illumination color compensation based on the finite-dimensional linear model, estimates the spectral power distribution of the illumination based on dichromatic reflection model and reconstructs surface reflectance functions from received color signals, based on which colors in the image are able to be recovered using the finite-dimensional model.In order to verify the effectiveness of color correction algorithm, a series of images taken under CIE A illuminant which are yellow-biased were processed to evaluate the performance. They were carefully chosen so that these images can be representative in the visible range. The original images were transformed to the images under the CIE Standard Illuminant D65. Color difference of recovered image and standard image as well as original images and standard image were calculated to evaluate the performance.With the illumination color compensation algorithm based on color space transformation, the color difference between the recovered and standard image declined by 68.98% compared with that between the original and the standard image. With illumination color compensation algorithm based on the finite-dimensional linear model, the color difference between the recovered and standard image declined by 74.34 compared with that between the original and the standard image. The experimental results show the effectiveness of the proposed methods.