块漂移引导的非局部纹理移除

发布时间：2018-01-22 20:30

本文关键词： 纹理移除保持结构块漂移非局部均值　出处：《计算机辅助设计与图形学学报》2017年09期 　论文类型：期刊论文

【摘要】：保持结构的图像滤波是诸多计算摄影学和图像分析应用的必要预处理方法.传统的纹理滤波方法仅仅考虑了图像的局部对比信息,当图像中包含对比度较强的纹理时,这些纹理会干扰现有方法,导致边界模糊等副作用.为此,提出一种基于块漂移的联合非局部均值滤波的方法.首先利用块漂移的方法选择出最能表示每个像素点纹理信息的图像块,生成具有判别性的引导图,该引导图在抑制纹理信息的同时保持住了原图的结构部分;然后借助该引导图进行非局部的联合滤波,通过考察图像块之间的相似性进一步降低由于引导图的模糊而引起的滤波后图像边界的模糊现象,从而得到清晰的图像结构信息.与其他经典方法进行实验的结果表明,该方法可以更好地移除纹理并且能够保持结构信息;此外,还将纹理移除应用到边缘检测和纹理增强,显示了保持结构的纹理移除有助于提高二者的效果.
[Abstract]:Image filtering with preserving structure is a necessary preprocessing method for many applications of computational photography and image analysis. The traditional texture filtering method only considers the local contrast information of the image. When the image contains strong contrast texture, these textures will interfere with the existing methods, resulting in edge blurring and other side effects. A joint nonlocal mean filtering method based on block drift is proposed. Firstly, the block drift method is used to select the image blocks that can best represent the texture information of each pixel, and then the discriminant guide graph is generated. The guide graph not only suppresses the texture information, but also keeps the structure part of the original image. Then, by using the guide map to carry out non-local joint filtering, by investigating the similarity between the image blocks, we can further reduce the fuzzy phenomenon of the filtered image boundary caused by the ambiguity of the guide image. The experimental results with other classical methods show that the method can remove texture better and keep the structure information. In addition, texture removal is applied to edge detection and texture enhancement, which shows that texture removal with preserving structure can improve the effect of both.
【作者单位】：大连理工大学数学科学学院;石家庄铁道大学信息科学与技术学院;东北石油大学数学与统计学院;
【基金】：国家自然科学基金(61363048,61402300,61572099) 中央高校基本科研业务费专项基金(DUT16QY02)
【分类号】：TP391.41
【正文快照】： 很多自然图像和艺术作品都包含了纹理,如草坪、十字绣以及瓷砖镶嵌艺术图等,如图1所示.在很多情况下,人们的视觉系统会忽视纹理图案,更关注图像的主要结构部分.然而,当人们通过细缝裁剪进行图像放缩,或者对图像进行细节增强、特征边提取以及图像合成等操作时,计算机容易将纹理

【相似文献】