基于显著图的可变模板形态学去雾方法

针对暗通道先验去雾中存在的光晕现象和天空区域颜色失真现象,提出了一种基于自适应可变形结构元(Adaptive deformable structuring element, ADSE)中值滤波结合灰度形态学重构精细化透射率的方法.该方法利用透射率与图像细纹理结构的无关性,由有雾图像的灰度图计算显著图(Salience map, SM),将SM作为导向图计算ADSE,用生成的ADSE对最小颜色通道图像进行自适应中值滤波运算;其次,以粗估计暗通道图像为标记图像,以自适应中值滤波后的图像作为模板图像进行灰度形态学重构运算,获得精细化暗通道图像,继而得到精细化透射率;最后,针对天空区域,引入最优化透射率方法,对天空和非天空区域做统一的运算得到最终透射率,完成图像去雾.本文算法对真实场景具有很好的去雾效果,同时,基于形态学的运算易于并行化和硬件实现.

Image Haze Removal with Salience Map and Morphological Adaptive Filtering

The dehaze image based on dark channel prior presents the phenomena of halo effect and color distortion in sky region. In this paper, the median filtering with adaptive deformable structuring element (ADSE) and the morphological reconstruction are introduced to estimate the fine transmission. The transmission does not relate to fine texture, so the median filtering can be performed on the minimum channel with ADSEs. The ADSEs can be computed by the salience map (SM) which is computed by haze image. Then, the filtered image and dark channel image of the haze image are used as the mask and marker images, respectively. The mask and marker images are used to perform morphological reconstruction for fine dark channel image and fine transmission. Finally, the transmissions in nonsky and sky regions are fused by the optimized medium transmission method. The experiment results show that the proposed method can obtain good dehazing effect, especially in real-world images. The proposed algorithm is based on morphology operations, which is easy for parallel computing and hardware implementation.