Single Image Defogging Method Based on optimized Double Dark Channel with Gaussian Weighting
Saisai Zhang, Yimin Tian, Liwen Shen, Hai Wang, Yunfei Du, Hongmei Chen · 2022
In this paper, we propose a single image defogging methods based on an optimized double dark channel with Gaussian weighting to address the problems of artefacts and residual fog at depth-of-field abrupt changes in the defogged images obtained by the traditional dark channel a priori algorithm. The two dark channels are first obtained using super pixel block filtering and median square filtering for each layer, the two filtered dark channels are then combined at the pixel level and enhanced by constructing a Gaussian weighting function, and then the transmittance is optimized using a guided filtering method. The fog map is then converted to HSV color space and the white areas containing the sky etc. are extracted, the probability function was introduced to take the average of the brightness components of the first 10% of the pixel points in the white area as the atmospheric light estimate. Finally, the contrast stretching method was used to improve the image brightness. The experimental results show that the proposed algorithm can better preserve the details of the image and remove the residual fog at the depth-of-field abrupt changes, and improve the artifacts with good visual effects.