Research And Realization Of Automatic White Balance Method Based On R/G-B/G Histogram

Phenomenon of color shift is a critical problem in digital image processing,which is caused by the fact that photographic equipment itself is unable to keep thecolor constancy. In the field of network monitoring, ensuring vivid and accurateimage is a basic requirement. Keeping white balance is one of necessary steps indigital image processing and it directly determines whether quality of the image isfine or not.So far, the available white balance algorithms are almost divided into twocategories, one is methods with supervision and the other is methods withoutsupervision. And the advantage of unsupervised white balance algorithm is itssimple computation and this algorithm is currently widely used in digital imageprocessing. These approaches calculate the color temperature of the light source bymaking a series of underlying assumptions while the supervised ones estimate thecolor temperature by machine learning. Compared in the correction results, thesupervised approaches work better than the unsupervised ones. The main reason isthat they have reserved plenty of prior knowledge and are not constrained by someassumptions. As mentioned above, the unsupervised algorithm is heavily related tothe assumptions and may end up in failure if the scene does not meet the presetconditions. Also, the main defect of supervised methods is the discrete sample spacesince they need a lot of experimental samples.In this dissertation, combining the simple computation in unsupervisedalgorithms and the well effect of the supervised ones, a new white balance method isdesigned and realized, which needs low computational complexity but is able toobtain relative satisfied correction results. This algorithm is based on the colorcorrelation theory in supervised white balance algorithms. The basic idea of thisalgorithm is to analyze distribution of the gray point in the rb histogram under thecondition of different color temperature so as to get positional relationship betweenthe gray points and color temperature. Then we can utilize the distribution of theimage gray points in the rb histogram under the condition of unknown colortemperature to estimate color temperature of the scene. Meanwhile, the probabilityof light source is introduced to correct the estimate value of color temperature. Asfor low illumination, the corresponding improvement solutions is proposed, makingup for the drawback that correction in low illumination tends to be yellowish.Compared in the experiments, the proposed algorithm reduces thecomputational complexity without loss of correction accuracy. What’s more, thisalgorithm covers the defect of discrete sample space as well as achieving satisfied correction results in low illumination.