Research Of Color Image Restoration In Foggy Weather

Cooling and water vapor full are formation conditions of fog. When the heat ofthe surface dissipates and the temperature has dropped to a certain extent, thehumidity of the air will increase and moisture will reach saturation. Excess moisturewill form small water droplets or ice crystals suspended in the air layer on the ground.When the visibility is below one kilometer, it is called fog. Small water droplets in thefog have effects of visible light absorption and scattering. Absorption causes energyattenuation of the incident light. Scattering is divided into the forward scattering andbackward scattering, which plays a major role in the scattering. Using a scatteringlayer transmission model proposed in this thesis and backscatter prior knowledge, wecan get the optical depth information of a fog image.This thesis analyzes reasons for the formation of fog and atmospheric scatteringmodel proposed by Narasimhan, and describes steps of removing fog in color image.After the fog image is transformed from RGB color space to HSV color space, theluminance image is abstracted and the atmospheric light information can be estimated.The optical depth information and the transmittance map can be obtained with thelayer transmission model. Meanwhile, the transmittance can be further optimized withthe guided filter and threshold method. Thus, the lightness component can be restoredwith the atmospheric light information, transmittance, and atmospheric scatteringmodel. After the image is transformed from RGB color space to HSI color space, thesaturation image can be abstracted and corrected. The restored lightness, correctedsaturation and hue of fog image can be as HSI color space of the correspondingcomponent, which can be converted to RGB color space. In conclusion, the fog imagerecovery result is obtained.In all kinds of fog algorithm, He Kaiming is more efficient. He uses a darkchannel prior to get the transmittance map, and optimizes it with the method of softmatting, but this algorithm needs a longer running time. This thesis uses thebackscatter prior knowledge to get the optical depth information, and then get thetransmittance, which can be optimized with the guided filter and threshold method.This optimization of transmittance takes shorter running time. After correction ofsaturation, the recovery of fog image has less color distortion. Compared with other classics defog algorithm, the defog algorithm proposed in this thesis has advantages ofless color distortion, shorter running time and better total effect.