Research On The Algorithms Of Single Image Haze Removal Based On Atmospheric Physics Model

Haze removal for degraded image is a fundamental and hot problem in computer vision, which finds a wide application in various areas. However, the reasons for haze image degradation are very complicated and the information in haze image itself is insufficient. Currently, the degraded process of a haze image can’t be described by any algorithms and models perfectly, and the previous algorithms are not good enough when using them in particular scene. There is still much works to improve the visual effect. Thus, it is necessary to research the haze-degraded image clearness techniques based on the analysis of image degradation mechanism.In this paper, image dehazing algorithm is divided into two types of image enhancement and atmospheric physics model, followed by a detailed description of the theoretical basis of atmospheric physics model, then describes the incident light attenuation model and atmospheric attenuation model, which essentially analyze haze image degradation mechanisms and causes.For the inadequacy of the dark channel prior, we improve the algorithm from two aspects: one is time complexity, using an image guided filter rather than soft matting to significantly reduce the amount of computation and improve the running time; the other is when the dark channel prior is invalid, redefine the restored image of solving equations, repair the image dehazing which has sky area or white objects, the comparative experiments from time and image quality evaluation.Traditional contrast enhancement algorithms do not dahaze from the nature, the article which is based on the atmospheric physics model proposes the optimization of contrast enhancement algorithms, formulate a cost function that consists of the contrast term and the information loss term, by minimizing the cost function, the proposed algorithm enhances the contrast and preserves the information optimally. Using an image guided filter optimizes the transmission map, and ultimately get no haze image. Experimental results show that the proposed algorithm effectively removes haze and is the extension of the dark channel prior.For the effect of the transmission map to restore the dehazing image, restore an image from the optimal transmission map. Firstly, the theoretic and heuristic border boundary of the transmission map is introduced. Then incorporate two scene priors model, including locally consistent scene radiance and context-aware scene transmission, turning haze removal problem into a constrained minimization problem, and finally solve it by quadratic programming. The derivation of the theretic border to verify the correctness of the dark channel prior statistical law, the proposed algorithm gets a more accurate transmission map, captures fine-grained boundaries and does not require post-processing. Finally, compare to the other advanced algorithms, experimental results show the robustness of the algorithm.