首页 | 本学科首页   官方微博 | 高级检索  
     

大尺寸纹理的实时合成
引用本文:陈 昕,王文成.大尺寸纹理的实时合成[J].软件学报,2009,20(Z1):193-201.
作者姓名:陈 昕  王文成
作者单位:中国科学院 软件研究所 计算机科学国家重点实验室,北京 100190;中国科学院 研究生院,北京 100049;中国科学院 软件研究所 计算机科学国家重点实验室,北京 100190
基金项目:Supported by National Natural Science Foundation of China under Grant Nos.60773026, 60873182, 60833007 (国家自然科学基金)
摘    要:提出一种纹理合成方法,可实时高质量地生成大纹理.它先基于纹理特征变化的周期性分析,得到合适的纹理块尺寸,以使所划分的纹理块能高效反映这种周期性变化,便于生成高质量的纹理;然后,它在目标纹理上均衡地分布纹理块,使得垂直方向和水平方向上相邻的纹理块之间都留有一个块尺寸大小的空白区域,再对空白区域进行填充,以完成目标纹理的生成.显然,布块操作和填充操作均可并行地进行.同时,为每个纹理块预先生成可与其邻接匹配的纹理块集合,以便在填充计算时可用简便的集合求交计算来进行邻域约束的搜索,并将这种求交计算放在CPU中进行,而将邻接纹理块在重叠区域的缝合计算放到GPU中进行,以综合利用CPU和GPU的优势.实验表明,新方法可在一般微机上以45帧/秒的速度高质量地实时合成1024*1024的大纹理,而这是已有技术难以达到的.

关 键 词:纹理合成  大尺寸纹理  实时  并行  GPU
收稿时间:2009/5/15 0:00:00
修稿时间:2009/7/23 0:00:00

Real-Time Synthesis of Large Textures
CHEN Xin and WANG Wen-Cheng.Real-Time Synthesis of Large Textures[J].Journal of Software,2009,20(Z1):193-201.
Authors:CHEN Xin and WANG Wen-Cheng
Abstract:This paper presents a novel technique for synthesizing large textures of high quality in real time. By analyzing the texture periodicity, patches in an optimized size are generated to well represent the variation of exemplar features. Then, during synthesizing, the paper first distributes patches on the output texture with a vacant region left between any pair of neighboring patches in every row and every column, where a vacant region is also in the same size as a patch. Thereafter, patches are selected to fill the vacant regions in the output texture. Obviously, both patch distributing and vacant region filling can be executed in parallel. To accelerate, for each patch, the paper constructs a set of matching patches that can be efficiently merged with the corresponding patch. The computation of patch selection for vacant regions, therefore, can be simplified into set intersection. Moreover, patch distribution and set intersection are performed on a CPU while patch stitching are executed on a GPU, to take advantage of both CPU and GPU. Experimental results show that the presented method is able to generate a high-quality texture in 1024*1024 pixels at over 45 frames per second, which is hard to achieve by existing techniques.
Keywords:texture synthesis  large texture  real-time  parallel  GPU
点击此处可从《软件学报》浏览原始摘要信息
点击此处可从《软件学报》下载全文
设为首页 | 免责声明 | 关于勤云 | 加入收藏

Copyright©北京勤云科技发展有限公司  京ICP备09084417号