Refined transmission map using guided filter
XiaoYuan Zhang, Meiping Shi, Tao Wu, Jian Li · 2015
In this paper, we propose a new approach to refine the estimated transmission map which based on dark channel prior. Traditionally, the raw transmission can be refined by guided filter using hazy image, but atomizing blurred salient edges and made transmission map including lots of redundant texture details which represented error information of depth. We obtain transmission map under the small and large scale mask using the minimum filtering. In order to have abundant high frequent information as well as accurate low frequent information from transmission. Firstly a new edge-preserving smoothing method named L0 smoothing will be conducted to diminish insignificant details in the transmission map under small scale mask, L0 smoothing can eliminate a manageable degree of low-amplitude structures and globally control the number of non-zero gradients. Then, we use guided filter based on a local linear model for information fusion of different frequent information. We demonstrate that the proposed algorithm can preserve prominent edge and keep the transmission map precise in a more appropriate area.