A Research Of Multi-illuminant Color Constancy Algorithm Based On Biological Visual Mechanism

Image processing is a very important orientation in the field of computer vision, while color bias correcting is a prerequisite for most image processing. Color constancy is a phenomenon in psychology that human visual perception of color does not change with the scene lights varied, however, it doesn’t work for image acquisition devices. The color constancy algorithm is to solve the problem that how to make equipment get rid of influence of light source itself, and show the inherent reflectivity of the scene.Most work about color constancy is based on the assumption that scene is illuminated by a single light, which means the wavelength of light concentrated in a very narrow range and can be considered a single light source. Since the assumption make research simple and easy, few people to study more complex situation in multi-illuminant. In fact, multi-illuminant is more universal, no matter indoors or outdoors under more than one light source usually. There have a few methods based on the assumption of multi-illuminant in color constancy. The methods are just extension or transformation by single-illuminant algorithm.In this thesis, we construct model by excavating the mathematical simulation of vision physiology. In the third chapter, we establish multi-illuminant color constancy algorithm model just using a high-luminance area of image based on the consequence of psychophysical experiments and the horizontal or ganglion cell receptive field mechanism. In the fourth chapter, we use ON and OFF type receptive fields to construct the algorithm which response to bright and dark regions of image. And it is accordant with human vision for processing light extraction and correction.To conclude, the image processing algorithms for multi-illuminant color constancy have achieved the purpose. For the majority of multi-illuminant image, it can be a good estimate for the position and color of the light source, and the corrected image is coordinating with the human visual feelings. By comparing with other algorithms and testing the existing image database, some achieve optimal results. That the model based on physiological mechanisms can be better applied to a specific image correcting and provide a new method for image processing.