基于截尾全变差Retinex算法的图像复原技术

Image Restoration Based on Truncated Total Variation for Retinex Algorithm

对照明环境提出全新假设,并在此基础上提出一种截尾全变差Retinex模型,通过分析具有明显边界的阴影特点,采用截尾全变差模型描述阴影特性,从而抑制照度非均匀对图像质量的影响,并与全变差、二阶全变差以及多尺度Retinex方法进行对比。对比实验表明,提出的方法可有效抑制传统模型在处理照明非均匀区域时产生的晕圈伪影,显著增强图像质量,对于具有强烈阴影的图像,所提算法仍能有效去除阴影,复原图像结构相似度与CIEDE2000值可达81.75%与22.352 3。

A new assumption with respect to non-uniform illumination is proposed. Based on that, a truncated total variation functional model for Retinex is proposed. Through analyzing the features of shadow of apparent boundary, the effects from the non-unifoml illumination and shadow are suppressed effectively by the proposed method. Experiments using simulated and real data are conducted to compare the performances between Total variation, HoTVL1, Muhiscale Retinex and proposed model. Results show that the proposed method suppresses the halo effect efficiently when dealing with non-uniform illumination parts and significantly improves the quality of the image. The proposed model can recover images with strong shadows, and SIMM and CIEDE2000 are 81.75% and 22.352 3 respectively.