Polarimetric dehazing for color images based on spectral analysis

Many polarimetric hazy image enhancing algorithms have been proven effective in recovering grayscale images, but when using these algorithms for RGB images, the by-product blue-shift comes along. It is found that light under short wavelength such as the blue light has less power to penetrate the hazy particles thus needed to be processed differently. In this paper, we analyze the cause of blue-shift for RGB images through spectral analysis, propose a novel polarimetric dehazing algorithm based on spectral analysis to solve that problem, and put forward an index to evaluate the amount of blue-shift of RGB images. Our method separates the dehazing process by spectral channels and processes each channel with different polarimetric dehazing parameters. The color dehazing result is achieved through weighted composition of the recovered images. In the experiment, a good dehazing performance with a good retention of color information is demonstrated, the amount of blue-shift can be lightened by at least 59%. Besides, our method is also proven to be useful in better recovering the detailed information and suppressing the noise.