位置:成果数据库 > 期刊 > 期刊详情页
基于Graph Cut的快速纹理合成算法
  • 期刊名称:计算机辅助设计与图形学学报
  • 时间:0
  • 分类:TP391.41[自动化与计算机技术—计算机应用技术;自动化与计算机技术—计算机科学与技术]
  • 作者机构:[1]华南理工大学计算机科学与工程学院,广州510641
  • 相关基金:国家自然科学基金(60573019);广东省自然科学基金(05103541);广东省自然科学基金博士科研启动基金(05300198,07300561).
  • 相关项目:图像分割与任意形状对象细分曲面重建算法的研究
中文摘要:

提出一种带边界匹配的基于Graph Cut的快速纹理合成算法.通过将纹理样本以不同的位移贴到输出图中完成合成,重叠区域的像素取值由Graph Cut确定.引入边界图辅助位移搜索,以增强合成结果的边界连续性;在预处理过程中计算2个相同样本在所有相对位移下的匹配误差,选取一部分误差最小的位移组成“优选位移”集合,合成过程中的块间相对位移仅从此集合中选取,大大地提高了合成速度.实验结果表明,该算法合成质量高、适用范围广、合成速度快.

英文摘要:

This paper proposes an efficient method technique and edge map. By the method, the synthesis is for texture synthesis based on graph cut made by iteratively pasting the sample to the output with different displacements, and graph cut technique is used to find the optimal patch boundaries. To make the edge continuously better preserved, edge map is introduced to measure the structural similarity in the displacement search. In the preprocessing step, match errors for all relative displacements of two samples are calculated, and several displacements with the smallest match errors are selected to form the "optimal displacement set". In the synthesis step, the relative displacements are fully chosen only from this set. Experimental results show that the synthesis process is greatly improved, and high quality synthesis results can be obtained by the algorithm for a variety of textures.

同期刊论文项目
同项目期刊论文