计算机科学 ›› 2016, Vol. 43 ›› Issue (4): 299-302.doi: 10.11896/j.issn.1002-137X.2016.04.061

• 图形图像与模式识别 • 上一篇    下一篇

基于GPU加速的保结构纹理合成

汤颖,林琦峰,肖廷哲,范菁   

  1. 浙江工业大学计算机科学与技术学院 杭州310023,浙江工业大学计算机科学与技术学院 杭州310023,浙江工业大学计算机科学与技术学院 杭州310023,浙江工业大学计算机科学与技术学院 杭州310023
  • 出版日期:2018-12-01 发布日期:2018-12-01
  • 基金资助:
    本文受国家自然科学基金(61003265,7),浙江省自然科学基金(LY14F020021)资助

GPU-based Texture Synthesis with Preserved Structures

TANG Ying, LIN Qi-feng, XIAO Ting-zhe and FAN Jing   

  • Online:2018-12-01 Published:2018-12-01

摘要: 提出一种基于Chamfer距离的保结构纹理合成方法。使用Chamfer距离度量纹理结构特征的相似度,在查找匹配块的同时计算纹理在颜色空间和结构特征空间的匹配度,从而解决以往纹理合成中有显著结构特征的纹理容易出现不连续的问题。但是Chamfer距离的计算量很大,而且随着纹理合成图分辨率的提高,计算成本会变得相当高昂以至于难以负担。因此,提出了基于GPU加速的保结构纹理合成方法,通过并行查找匹配块提高合成效率。实验证明本方法既提高了结构纹理的合成质量,又使得保结构纹理方法的合成速度大大提高,且与纹理合成图的分辨率无关。

关键词: 纹理合成,Chamfer距离,结构特征,GPU

Abstract: We proposed a texture synthesis method based on Chamfer distance,which preserves the texture structures.We measured the similarity between texture structures by Chamfer distance and computed the distances of textures in both color space and structure feature space during the search of the matching texture patches.In this way,we solved the problem of discontinuity in synthesized textures with obvious structural patterns.However,the computational cost of Chamfer distance is very expensive,and it will become unaffordable as the resolution of the synthesized textures increases.To address this problem,we further presented a GPU-based algorithm to accelerate the computation of Chamfer distance during texture synthesis.The matching patches are searched in parallel on GPU to greatly improve the computational efficiency.Experimental results show that our method improves the quality of structured texture synthesis and accelerates the synthesis speed,which is irrelevant to the resolution of synthesized textures.

Key words: Texture synthesis,Chamfer distance,Structure feature,GPU

[1] Holland J.Adaptation in Natural and Ritificial System [M].Arbor.Univ.of Michigan Press,1975:17-24
[2] Zhou M,Sun S D.Theories and applications of genetic algorithm[M].Beijing:National Defense Industry Press,1999:18-24(in Chinese) 周明,孙树栋.遗传算法原理及应用[M].北京:国防工业出版社,1999:18-24
[3] Wang X P,Cao L M.Theories,applications and software implementation[M].Xi’an:Xi’an Jiaotong University Press,2001:68-76(in Chinese) 王小平,曹立明.遗传算法——理论、应用与软件实现[M].西安:西安交通大学出版社,2001:68-76
[4] Cui Y D.Determine the optimum section length to cut form arectangular sheet to generate cutting patterns for rectangles of a single size[J].Computer Engineering,2004,0(7):178-180(in Chinese) 崔耀东.长板单一尺寸矩形毛坯定长分割优化排样[J].计算机工程,2004,30(7):178-180
[5] Cui Y D,Huang J M,Zhang X Q.Recursive algorithm for unconstrained two-dimensional guillotine cutting problem rectangular pieces[J].Journal of Computer Aided Design and Graphics,2006,18(7):948-951(in Chinese) 崔耀东,黄建民,张显全.矩形毛料无约束二维剪切排样的递归算法[J].计算机辅助设计与图形学学报,2006,18(7):948-951
[6] Yan X.Research of rectangular three stages strip packing problem for genetic algorithm[D].Guilin:Guangxi Normal University,2009:15-17(in Chinese) 严玄.矩形件三阶段带排样问题的遗传算法的研究[D].桂林:广西师范大学,2009:15-17
[7] Yao Y.Heuristics for rectangular strip packing problem basedon hierarchical search filled strategy[J].Engineering Journal of Wuhan University,2014,47(6):854-858(in Chinese) 姚怡.采用分层搜索填充策略的启发式带排样算法[J].武汉大学学报,2014,47(6):854-858
[8] Zhao X F.Genetic algorithm of solving the rectangular piecesstrip packing problem[D].Guilin:Guangxi Normal University,2008:15-18(in Chinese) 赵新芳.解决矩形件带排样问题的一种遗传算法[D].桂林:广西师范大学,2008:15-18
[9] Jiang X B,Lv X Q,Liu C C.Improved genetic algorithm for the rectangular pieces packing optimization[J].Computer Enginee-ring and Application,2008,4(22):244-248(in Chinese) 蒋兴波,吕肖庆,刘成城.一种用于矩形排样优化的改进遗传算法[J].计算机工程与应用,2008,44(22):244-248
[10] Jin J,Su Y.Improved adaptive genetic algorithm[J].Computer Engineering and Application,2005,41(18):64-69(in Chinese) 金晶,苏勇.一种改进的自适应遗传算法[J].计算机工程与应用,2005,41(18):64-69
[11] Mennon A,Mehrotra K,Mohan C K,et al.Characterization of a class of sigmoid functions with applications to neural networks[J].Neural Networks,1996,9:819-835

No related articles found!
Viewed
Full text


Abstract

Cited

  Shared   
  Discussed   
No Suggested Reading articles found!