Image Dehazing Algorithms Using Dark Channel Prior Based On Transmittance Correction

Due to the impact of atmospheric scattering, images obtained in conditions of fog usually have poor contrast, obvious, partially details missing, and color distortion. It seriously affects the value of image and directly limits the normal operation of outdoor vision systems such as intelligent navigation, highway monitoring, satellite remote sensing systems. Moreover, foggy day has an impact on travel and work for people. This thesis mainly studies the haze removal from single image based on dark channel prior. The main contents are as follows:(1) The bright areas like the sky are not meet the dark channel prior, so color distortion occurs in the region after dehazing. multi-threshold detection of sky and a way to correct transmission rate is came up with to solve the problem. First, by analysis and summary characteristic of the sky region, three threshold, including the brightness threshold, the details threshold, and the transmittance threshold is provided to detect the sky area. Then, in order to obtain a more accurate estimate of transmission in sky region, a tolerance mechanism is used to correct the transmission. Experimental results show that the algorithm not only can accurately detect the sky region, but also color distortion in the sky area is effectively avoid by correcting the transmission.(2)To avoid blurring the depth information when refine transmission by existing algorithms, detection of depth mutation area is innovatively came up with, and then in order to maintain depth information, minimum filter is applied to the region. The algorithm take advantage of the dark channel prior can be roughly representative of picture’s depth information. Firstly, the depth mutation region is found by making a subtraction of two different size templates of dark channel. Then, the minimum filter is applied to maintain the depth information. Simulation results show that the algorithm can better maintain the depth information, and the fog of depth mutation area can be removed more thoroughly.(3)The way above has low speed, to address this problem, double-regional filter is used to optimize the transmission. Firstly, Weighted median filter is used to achieve rapid demisting.While the result of defogging based on median filter is a little dark, in order to improve result of defogging, then the pixel values of filtered image is reduced by a correction factor. Finally, double-regional filter, that is median filter combine with minimum filter is used to suppress black spot phenomenon. Experimental results show that the algorithm has improved much in speed. And compared with the result of general median filter directly, our algorithm can improve the results of dim phenomenon with more natural color, more details, and can suppress dark spots of leaf nodes or other small crevices.