基于导向滤波优化的自然图像去雾新方法

Improved Natural Image Dehazing Algorithm Based on Guided Filtering

由于暗通道先验的单幅图像去雾方法通常使用导向滤波器来优化大气透射率,在保持较好去雾效果的同时能大大降低算法复杂度。但是由于采用有雾图像作为导向滤波器的导向图,使得优化后的透射图在景深相同或相似的区域不够平滑,并且包含大量的细节信息,导致去雾后的图像在该区域的可视性差。因此,提出了一种新的基于导向滤波优化的单幅图像去雾方法。该方法利用基于大气光幕的导向图对大气透射率进行导向滤波,将优化后的大气透射率结合大气物理散射模型恢复出无雾图像。实验结果表明,相比暗通道先验方法,该方法恢复出的无雾图像色彩更加真实,细节更加丰富,结构更加清晰。

The guided filter is often used in single image dehazing algorithm which is based on dark channel prior.This strategy can optimize atmospheric transmittance, preserve impressive dehazing effects as well as reducing the complexity of algorithms. However, the optimized transmittance map is not smooth enough and contains abundant detail information when adopting the hazy image as the guidance of guided filter, which will result in the relative poor visibility of the recovered image undoubtedly. Thus, this paper proposes an improved single image dehazing algorithm based on optimized guided filtering. This algorithm utilizes the atmospheric veil-based guide image to filter the atmospheric transmittance, then combines the optimized atmospheric transmittance and the atmosphere physical scattering model to recover the hazeoff image. Exhaustive experimental results on a variety of hazy images demonstrate that the color, details and structure of the recovered images are more powerful when compared with the dark channel prior method.