Research On Computational Color Constancy

Color is a very important and useful feature in the computer vision community. However, the image pixel values of an object could vary if the lighting conditions change, since any color imaging device’s response to light from imaged scenes depends on three factors: the nature of the illumination incident on the objects, the underlying physical property of the objects, and the sensor sensitivity of the imaging system itself. Therefore, as the urgent demands and challenges for computer vision applications continue to grow,accurate reproduction of the object’s color becomes a more critical issue. Color constancy theory can solve this problem. In recent years, a growing field of multispecular images has interested many scholars because they can be used in high quality image applications, and reconstruction of spectral reflectance is a new research direction of color constancy.Firstly, the paper provides the analysis of color theory and color constancy. Secondly, the new illuminant estimating algorithm based on dichromatic reflection model is raised for the images which have highlights. Fuzzy clustering and linear fitting are used to simplify the calculation of intersection point or slope of straight line. There is no limit on the number, color and texture of the highlight areas.Thirdly, a new algorithm on the basis of Gaussian Process Regression is proposed to fitting the relation between the image color and the illuminant.Experimental results show that the algorithm can effectively overcome the defects of the neural network and SVM. And finally the paper discusses the feasibility of increasing the number of dimension from 3-bands to 6-bands or 9-bands by adding filters in front of the lens. Then an idea of Reconstruction of Spectral Reflectance based on Supervised Learning is proposed.