Research On And Application Of Image Dehazing Algorithms

With the development of information technology, visual effect of images has been taking more and more attention., which is undoubtedly a huge challenge for pictures with haze or fog. The pictures has brought great inconvenience and safety problems for traffic, aerial photography, video surveillance and vehicle navigation. Therefore, the research on image dehazing technology is of great significance and urgency. The Dark Channel Prior method proposed by Dr. He Kaiming is a major breakthrough in the art and is also the focus of this study. The DCP algorithm is simple, effective, practical of the characteristics and automatic,it also has some drawbacks, such as dependence of the atmosphere scattering model, the choice of the filtering size, lack of immunity against the region of sk y and the dim of recovery image.On the basis of extensive study and research on image dehazing tec hnology and lots of experiments and innovative tries on DCP algorithm, this study has achieved good results, which are mainly in the following four aspects. First, we improved an color retention capacity evaluation method, based on which we proposed a comprehensive evaluation based on no reference evaluation. Second, for the choice of the size of filtering template, the author tried a large template filtering to the contrary and achieved good improvement effect. O n this basis, we proposed an adaptive filter ing method based on a large template, which can improve structural information retention and enhance the contrast at the cost of lost of details slightly. This method can be applied to the situations which do not need sufficient details. Third, for the estimation of atmospheric light, the author proposed a simple method of judgment on pixels with high brightness so as to exclude the interference of the region of sky, which is more simple and takes less time than the common and sophisticated method of segmenting the region of sky. This method can improve the restoration effect and provides a significant improvement for the common region of sky, but provides no improvement for the special region of sky with a color of khaki. Fourth, for the estimation of the transmission matrix, in order to overcome the issue of render color in the region of sky, the author proposed a preprocessing method before we got the initial transmission matrix, which is very different from the common methods which deal with the refinement of the transfer matrix. The recovery images obtain a higher contrast and greatly improves the retention of the structure, but it also will make the white clouds in the sky disappeared. Results and figures fully show that these methods greatly improved the DCP method, but they also have some shortcomings and are worth further study.