Image Haze Removal Improved Algorithm Using Dark Channel Prior

In bad weather conditions such as the fog or haze, due to the large number of small waterdroplets suspended in the atmosphere and aerosol, color of image has distortion and contrast andvisibility of image reduce, which has a direct impact on security of many areas such as urbantransportation, shipping, air traffic. In addition in the field of image processing many algorithmsassume that the input image or video was shot in the ideal weather conditions, so the image hazeremoval technology is particularly important.In recent years more and more new image hazeremoval algorithms have been proposed.Image haze removal algorithm is now mainly divided into two categories: fog imageenhancement and fog image restoration. The purpose of fog image enhancement and fog imagerestoration is to improve the quality of the image, such as increasing the sharpness of the image,increasing the contrast of the image, improving image visual effect. Paper focus on the algorithmthat is single image haze removal based on dark channel prior proposed by He-kaiming. Thealgorithm can obtain good effect for the majority of fog image, but time complexity, spacecomplexity of the algorithm is relatively high, and its efficiency is too low; besides it can not satisfythe real-time requirements well. For this, this paper puts forward the improvement.On one hand, thedark channel is got with sampling and interpolation; on the other hand,transmission rate isoptimized by using the minimum filtering instead of soft Matting algorithm.Besides, for the pixelpoints close to the atmospheric light intensity defog degree is reduced to avoid noise excessiveamplification, improve image visual effect.If using Dark channel prior to remove fog, image is relatively dim. To solve this problem, anew image enhancement algorithm is proposed in this paper. Firstly, the image enhancement curveis used for its brightness enhancement and brightness gain, then the three color channels value isincreased by the same proportion, but its each component proportion did not change, but its eachcomponent proportion did not change, so the image color did not change, but its brightness isimproved.To further illustrate the effectiveness of the algorithm proposed in this paper, this papercompare it with existing advanced algorithm, and analysis the advantages and disadvantages ofvarious methods.