Research On Single Image Dehazing Method Based On Dark Channel Prior

In the field of image processing, restoring the blurred image into clear is called image defogging. In the image acquisition process, the medium in the air such as air, vapor and suspended particles etc will scatter the light before light reaches lightsensitive components. Then, when the light reaches, the interference of these external factors has been reduced or even disappeared. Light in the air can also cause interference to the components which makes image blurred and detail information lost. In real life, the application of image defogging is widely used for outdoor surveillance systems, traffic systems, security systems in shopping malls and companies, etc. Without image defogging techniques, images in the above systems will be blurred and indistinct in the dark night or harsh environment, which greatly affects the working efficiency and people's lives.With the rapid development of computer technology, the image defogging techniques have been greatly improved. In the earlier time, by studying differences of the same scene under different conditions, we can achieve the purpose of the defogging, which is called multi image defogging. But it exists a biggest problem. In the image acquisition process, there are lots of interference factors. These factors are often unknown, only by keeping the different conditions can't ensure other factors in the same pace, and it is very difficult to achieve in the actual operation of the process. That is to say it's hard to obtain multi image of the same scene under different conditions. In addition to the actual operating difficulties, for a lot of real-time requirements of the system, the multi-image to the fog algorithm does not apply. That is why single image defogging techniques get rapid development.After searching references, we can summary the defogging algorithms of a single image at first. The algorithms can be simply classified into two types. One is called defogging algorithm based on image strengthening which is aimed to strength useful information. Fog algorithm based on image enhancement is through the fog image transform data and selectively highlight image valuable information or inhibit the useless image even interference information indirectly outstanding valuable information. The image may be different from the real object,but it must meet human visual needs. The main methods are histogram equalization, Retinex algorithm and so on. Image restoration based on the principle of the fog is just the opposite, it is through the analysis of the impact factors of the image acquisition process. Then set up physical models and inverse physical models to achieve the goals of defogging. The main method is algorithms of image defogging based on atmospheric scattering models. By analyzing the classic image defogging algorithm used for enhancing and restoring images, we choose the dark channel prior defogging algorithm based on atmospheric scattering models as a researching direction.The dark-channel prior algorithm is through a large number of clear image experiments, according to the characteristics of clear image statistics that objective law. The theory that in any clear image, in the absence of the sky and the bright object in the presence of the range is very small in the local area, there is always such a pixel, in RGB color model in a certain color channels in the strength value is very low, close to zero, known as the dark-channel pixel. The atmospheric scattering model and darkchannel prior theory can be used to calculate a rough transmittance value, good results can be obtained by optimizing the soft-matting. Although this algorithm can achieve good defogging effect, but due to the high computational complexity of soft matting, in the actual process takes a long time, it is not conducive to practical use.By analyzing the entire process of the dark channel prior defogging algorithm, we will point out the deficiencies and find out internal cause, target an optimization algorithm. Underestimate the boundary pixel leads to the halo phenomenon through analyzing the reasons. Finding the edge by edge detection, optimize the transmission rate by the maximum value of neighbor pixels in the edge instead of the edge pixel. Through the analysis on the valuation of atmospheric optical, we can know when the high brightness of the object in the scene, the dark-channel prior algorithm valuation is not accurate enough. So we find out the location of the light in atmosphere by setting a threshold value and take the maximum luminance value as the value of the light in atmosphere. Experiments have proved that the halo phenomenon do good removal effect and greatly shorten the time to improve the efficiency.