Single Image Dehazing Based On A Improved Tolerance Algorithm

Images captured by video collection equipment under haze conditions usually have low identification, which makes the performance of intelligent transportation, surveillance, and target identification system reduced. Therefore the enhancement and dehazing of haze images play a very important role.Firstly,the algorithm of the single image haze removal using dark channel prior is introduced in detail.Transmission of the sky area,which are not matched the assumptions of dark channel prior,will be compensated through adding tolerance mechanism model in the dehazing model and the estimation of the error transmission would be corrected to accurate one. A general method for single image dehazing is developed based on tolerance mechanism model. This paper describes an improved tolerance mechanism, which combines the power function and logarithmic function to form a new compound function. Experiment results validate the performance of the proposed approach which makes the color of sky area is natural. The results show that many feature hided in haze are visible and the serious distortion of color in the sky area become natural.Then, the parameter K of the tolerance mechanism, a kind of statistics of transmittance which can recover haze-image to free-image without distortion will be optimized. A optimized K value would be obtained with calculating statistic data of the different range of the minimum transmission and the different thickness of the haze-image. A appropriate parameter K of the tolerance for different image is acquired depended on the minimum transmission ranges and the improved tolerance mechanism. Experimental results show that the proposed method can get good effect for different images and overcome the limitation of the previous method. We can automatically estimate the effective value of the tolerance parameter K by the proposed method, which recover the free-image without color distortion from the haze-image with readjusting flexibly the value of the tolerance parameter K.Finally, more comparisons between our results and the original methods are presented to illustrate the effectiveness of the proposed algorithm.