基于双边核回归的相对约减纹理分解方法  

Related Reductive Texture Decomposition via Bilateral Kernel Regression

作　　者：苏卓[1,2] 吴学标[1] 曾碧怡 颜吉超 罗笑南 Su Zhuo Wu Xuebiao Zeng Biyi Yan Jichao Luo Xiaonan(School of Data and Computer Science, Sun Yat-sen University, Guangzhou 510006 National Engineering Research Center of Digital Life, Sun Yat-sen University, Guangzhou 510006 Beijing Key Laboratory of Multimedia and Intelligent Software Technology, College of Metropolitan Transportation, Beijing University of Technology,Beijing 100124 Research Institute of Sun Yat-sen University in Shenzhen, Shenzhen 518057)

机构地区：[1]中山大学数据科学与计算机学院,广州510006 [2]中山大学国家数字家庭工程技术研究中心,广州510006 [3]北京工业大学城市交通学院多媒体与智能软件技术北京市重点实验室,北京100124 [4]中山大学深圳研究院,深圳518057

出　　处：《计算机辅助设计与图形学学报》2016年第12期2202-2209,共8页Journal of Computer-Aided Design & Computer Graphics

基　　金：国家"九七三"重点基础研究发展计划项目(2013CB329505);国家自然科学基金(61502541;61320106008);广东省自然科学基金-博士启动项目(2016A030310202);中央高校基本科研业务费专项资金(中山大学青年教师培育项目)

摘　　要：图像中结构被纹理覆盖的现象无处不在,将结构从复杂的纹理中提取出来是一件非常有挑战但又极其重要的事情.为此,提出基于双边核回归的纹理分解方法,通过局部全变分的核描述子能将结构和纹理很好地区分开来,并将结果与双边核回归框架相结合:采用相对约减纹理分解来构造结构核描述子;再将该描述子与双边核回归融合来获得期望的结构感知滤波输出;最后采用一个稳定近似迭代的流程来实现所提算法.实验证明,与其他边缘感知滤波方法相比,文中方法能获得更佳的结构保持效果,并能被推广到多个视觉相关的应用上,如高动态范围色调映射、超像素分割等.It is ubiquitous that meaningful structures are appear over textured surfaces. Extracting them underthe complication of texture patterns is very challenging, but of great practical importance. Consequently,we have proposed a novel structure-aware filter via bilateral kernel regression with a variational structure-kernel descriptor which can extract main structures from textures through variational structure-kerneldescriptor and incorporate its results into the bilateral kernel regression： we first apply the related reductivetexture decomposition to construct the structure-kernel descriptor. Then, we incorporate the descriptor intothe bilateral kernel regression to achieve an expected structure-preserving output. Algorithmically, we proposea numerically stable approximation iterative procedure to achieve effective implementation. At last,some experimental results are presented to demonstrate that our approach leads to better or comparable performanceand is effect in some applications, such as HDR, super-pixel segmentation and so on.

关 键 词：边缘感知滤波 边缘保持平滑 结构纹理分解 

分 类 号：TP391.41[自动化与计算机技术—计算机应用技术]