Improved Dual Color Space Dark Channel Prior for Single Image De-Hazing

Abstract

Analyzing the environment and perform autonomous tasks such as identifying or recognizing the object, or navigating the route, the perceptual quality of outdoor scene image is important. One of the hurdles faced to carry out these automated tasks is fog or haze in the image caused due to scattering and absorption of light. To improve the quality of imaging systems, many methods have been devised to remove haze by researchers. One of the method used for dehazing image is to reverse physical of Atmospheric Scattering. Dark Channel Prior is an assumption which when applied on hazy image, Atmospheric Light and Transmission map can be approximated. DCP is implemented on RGB channel as well as on YCbCr channel to get intermediate haze free images. This thesis proposes novel transmission map fusion technique which speeds up the process of dehazing. Moreover, for color correction and improved quality another transmission map has been proposed based upon minimum pixel of RGB color space. This haze free image is then analyzed using state-of-the-art Blind Image Quality Indexes like Blind image quality assessment through anisotropy (BIQAA), Blind Image Quality Index (BIQI), Natural Image Quality Evaluator (NIQE), Blind non-Reference Image Spatial Quality Evaluator (BRISQUE) and Referenced Image Quality Indexes like Structure Similarity Index Measure (SSIM), Fog Effect based on Mean Square Error (FE-MSE). Evaluation results suggest that proposed technique gives better result than previously proposed techniques.