基于RTV模型图像分解的去雾算法

Image Dehazing Method Based on Image Decomposition of RTV Model

针对暗通道先验去雾算法出现的纹理细节丢失、边缘模糊、天空与明亮区域失真等问题,提出了一种基于RTV模型图像分解的去雾算法。首先利用相对全变分(RTV)模型将有雾图像分解为结构层与纹理层,然后建立一个指示纹理区域的遮罩对纹理层进行优化来解决纹理细节丢失的问题。为了减少结构层边缘的深度跳跃现象,建立了加权L1正则化模型对初始透射率进行优化。同时针对天空与明亮区域出现的失真现象,引入容差机制优化了该区域的透射率。最后将优化后的纹理层与去雾后的结构层重组得到最终的复原图像。实验结果表明,该算法复原后的图像在平均梯度以及边缘强度等客观评价指标上均好于其他几种对比算法,基本达到了纹理细节突出,边界清晰,饱和度适中的处理效果。

In this paper,a defogging algorithm based on image decomposition of RTV model is proposed to solve the problems of losing details,edge blur,sky and bright region distortion in the traditional dark channel prior defogging method.In order to solve the problem of missing texture details,the fog image is decomposed into a structure layer and a texture layer by using relative total variation,then a mask indicating the texture area is created to optimize the texture layer.In order to reduce the depth jump phenomenon at the edge of the structural layer,a weighted L1 regularization model is established to optimize the initial transmittance.At the same time,the distortion of the region is corrected by the tolerance K for the distortion phenomenon occurring in the sky and the bright region.Finally,the optimized texture layer and the defogged structural layer are recombined to obtain the final restored image.The experimental results show that the reconstructed image is higher than other comparison algorithms in terms of average gradient and edge intensity;and the effect of outstanding detail,clear boundary and moderate saturation are basically achieved.