Research And Realization Of Auto White Balance Algorithm Based On Color Temperature Estimation

Human visual system has the characteristic of color constancy, so the image that the human eye recepts will not be affected by the light source. CMOS image sensor’s work mimics human visual system, but it doesn’t have the color constancy characteristic, so the auto white balance function is necessary for the image sensor to eliminate the impact of the light source.Based on the research of the existing auto white balance algorithms, in order to settle the failure of the gray world algorithm when the color in the image is too little, and also to resolve the accuracy problem induced by the color space conversation, a new algorithm that based on color temperature estimation which processes in the RGB domain is applied in this paper. The R/G and B/G distribution of different light resources has to be researched as the groundwork, and the look up table of the familiar light resources is set. During the realization of the algorithm, the light resource of the current pixel is estimated based on the pixel’s raw R/G and B/G, and the pixels are dealt one after one. When the whole frame is finished, the color temperature that has the most number of pixels is determined as the light resource. Then R, G, B gain is calculated using the pixels in that color temperature; finally the whole image is adjusted according to the elicited gain, and the white balance function is realized.In order to measure the result of the algorithms, the average hue of the images is employed as the criterion. In this paper, the Matlab model and FPGA test platform is built to test the algorithm, and finally the dc synthesis is realized, and the working performance of the chip is tested. The test result shows that when the color in the image is abundance enough ,the color temperature based algorithm is little better than gray world algorithm ,the average hue of the image is reduced by 19.2%; when the image is occupied by a few main color blocks,the average hue of the image is reduced by 42.8%, the failure that the gray world algorithm faced is avoided effectively; Compared with the procession in other color domains, the result of this algorithm is better, and the precision errors introduced by the conversation between the color domains is eliminated. This algorithm has already been employed in two different chips, one has a 640*480 pixel array, and the other has a320*240 pixel array. Both of them work stably, and meet the requirement of most of the environments.