基于结构相似性感知的引导图像滤波

Guided Image Filtering Based on Structure Similarity Aware

借助引导图像对滤波过程进行指导,引导滤波器取得了较为理想的滤波输出,并在图像处理领域引起广泛关注.然而,现有滤波算法,如联合双边滤波,引导图像滤波等往往忽略图像间的结构差异性,从而导致其滤波输出容易产生纹理复制的缺陷.为解决这一问题,本文首先提出一个能够精确反映两图像结构关系的新指标,即结构相似性感知.继而在此基础上设计出一个新的引导滤波改进方法.相对现有滤波算法,该方法能够根据图像间的相似性自适应选择引导图像结构,从而有效避免了纹理复制的缺陷.

With help of the guidance image to direct the filtering process,Guided filters have achieved powerful performance and attracted more and more attentions in image processing area.However,due to ignoring the structure difference between the guidance image and the target one,existing methods,such as joint bilateral filtering and guided image filtering,usually suffer from texture copy in their filtering outputs.To overcome this drawback,this paper first proposes a novel indicator to describe the structure relationship between the two images more accurately,i.e.,structure similarity aware.Go a step further,a novel guided filtering method is proposed.In a nutshell,the proposed method can adaptively select the consistent structures in the guidance to supervise the filtering operation,and thus avoid texture copy effectively.