Research And Improvement Of White Balance Algorithms Of Digital Images

White balance algorithms are widely used in camera and video systems. Originally it is referred to as reinstalling the whiteness of a white object in a digital picture, which means wiping off the affect of an unknown illuminant on the color of the object and restore the color as if the picture was taken under a canonical illuminant. Therefore, white balance is also referred to as color constancy. This thesis investigated some of the main algorithms and make improvements on some of them.Researchers have developed a lot of White Balance Algorithms. Unfortunately, none of them could be applied well to all digital images, which means each algorithm has its own limits. The goal of this thesis is to find ways to eliminate some of the drawbacks of existing algorithms. The author focuses on two aspects: One is to develop new algorithms based on GW and WP that deal well with pictures once suitable only to one of these two. The other aspect is to develop a new algorithm to take off the edge pixel effect of ACE (Automatic Color Equalization), called MACE (Modified Automatic Color Equalization).In the first aspect, 5 sub-algorithms categorized into two groups were developed in this thesis. The end of one group of algorithms is to automatically pick up a better algorithm in GW and WP for a certain photo, which was defined as unsupervised algorithms. Since there are quite many versions of GW and WP that uses different color space, I developed distinct algorithms that can handles not only GW and WP that use the same color space, for instance, RGB, but also GW and WP that uses different color spaces, for instance, RGB for GW and YCbCr for WP. The other group of algorithm deal with the case in which GW and WP use the same color space but neither is suitable for the target photo. In this group of algorithms, I combine the normalization ratio of GW and WP into a whole new normalization ratio, and then use the new normalization ratio to calculate the color map of the target photo, which yield a better result than either GW or WP. According to sufficient experiments, these algorithms have reached their original purpose and are proved to perform better than the original algorithm.In the other aspect, the author developed a new algorithm called MACE to eliminate the edge pixel effect of ACE. I add two processes: the first is spatial windowing and the second is odd pixel elimination. With these two processes, MACE gets rid off the pixels that are irrelevant in the calculation of intermediate matrix R, thus removes the edge pixel effect. Through a lot of experiments, I find that MACE is highly effective in eliminating edge pixel effect.There are two aspects of improvements based on existing white balance algorithms developed in this paper. Both aspects not only broaden the applicable range but also improved the processing effects of original algorithms.