Research On Fast Image Defogging And Visibility Restoration

The images of outdoor scenes obtained in haze, fog and other weatherphenomena are usually have poor contrast and color fidelity. It must be difficult toestimate the objective using these blurred images, and it directly limited the outdoorvision systems. Therefore, in order to effectively improve the fog-degraded imagequality, reduce the effect of the fog, haze and other weather conditions to outdoorimaging system, there ispractical and theoretical meaning of the images defoggingand visibility Restoration fast and efficientlyThis dissertation for the image degradation in fog and haze, we detailedanalyzed the image degradation causes and fuzzy mechanism, researched on the basictheories and key technologies of defogging method, and processed the fog-images intwo aspects of image enhancement and image restoration. This dissertation madesome meaningful work for improving the existing defogging method and introducingnew ideas. The research contents of this dissertation are as followings:1. The Retinex image enhancement method is emphatic studied in the non-modelmethods. A new Retinex algorithm is proposed in order to solve the immanent defectsof the traditional Retinex algorithm,such as high complexity, the great calculation andthe halo phenomenon. It obtained preferable defog effect by improving the local andglobal image contrast, and the computation time is20%of the multi-scale Retinexalgorithm.2. The single image haze removal method based on the physical model ofhazy-image is studied in defogging method based on image restoration. After detailedanalyzing the high complexity of the defogging methodusing dark-channel prior, animproved and fast image defogging method is proposed. On the one hand, by using the bilateral filtering algorithm to refine the transmission map instead of the softmatting, it greatly reduced the complexity, and the consuming time is15%of theoriginal algorithm. On the other hand, for the dark-channel prior invalid scene,weadvanced an improved method to estimate the atmosphere light,it broadens theapplication of the dark-channel prior in a certain extent. The results of experiment ona variety of outdoor haze images demonstrated that the method can fast and efficientlyenhance the contrast and color definition of hazy degraded images, and greatlyimproved the visibility of the outdoor visual system in fog and haze.3. An image defogging method based on the global dark-channel prior isproposed to meet the real-time requirements of engineering. From the viewpoints of imagerestoration and image enhancement, the optimized models of global haze removal andself-adapting contract extending are established. It can directly recover a high quality haze-freeimage and obtained satisfactory visual effect. The contrast, detail and clarity of theimages had been obviously improved, and restored the vivid, true color of the scene.Mostly, the method greatly improved the rate of the original algorithm, and theexperimental results showed that the computation time of global dark-channel prioralgorithm is2%~5%of the original algorithm, it can meet the real-time requirementsof the general engineering.4. Since the brightness of image after haze removal is lower than the actual scene,the color is dim, and in the local area the haze is not completely removed, a simpleand effective method of image contrast and brightness enhancement is proposed. Itadaptively enhanced the image brightness, and modified the local contrast which isdue to the thick fog and haze.5. Combining the subjective and objective image quality evaluation, thisdissertation evaluated and analyze the experimental results image from theefficiency, the stability, obvious effect and universal application of the algorithm, aswell as contrast, entropy and the average gradient. The results, time statistics, as wellas the evaluation value is presented.It compared the defogging effect, the time, andevaluation value too.6. For the engineering applications, combing the theory of global dark-channelprior algorithm, a high performance image acquisition, enhancement, haze removaland transmission system was designed. It built a hardware platform based on FPGA,completed the algorithm implementing in hardware. The simulation result of timingsequence and the actual testing result were given.