同义图像融合系统设计与优化  

Design and Optimization of Synonymous Image Cloning System

作　　者：王宇新[1] 陆国际[1] 郭禾[1] 何昌钦[1] 杨元生[1] 

机构地区：[1]大连理工大学计算机科学与工程系,大连116023

出　　处：《计算机科学》2010年第8期283-286,共4页Computer Science

摘　　要：图像无缝融合技术是图像编辑领域的重要方向,传统的方法是将目标物体从源图像中分离出来后直接嵌入到目标图像中,而未考虑源图像和目标图像在语义上是否匹配。实现了一种新的图像融合系统,它用Gist特征描述图像的场景语义,并将场景语义匹配应用于无缝融合之中,使图像融合更具有现实意义,该系统称之为同义图像融合系统。鉴于语义匹配时间复杂度较大,采用异构多核环境下CUDA并行编程和OpenMP多核多线程方法进行优化,有效提高了系统的整体性能。实验表明,该系统不仅用户参与度较高,而且实用性较强,可以达到用户满意的融合速度和效果。Image seamless cloning technique is an important filed in image editing region. In previous work, researchers separated the target object from source image and then embed it into the target image without consideration of whether the source image and the target one are semantic matching. A novel and complete image cloning system was presented, in which Gist descriptors were used to describe image＇s scene semantics and semantic matching technique was applied to arbitrary cloning method to give it a realistic significance,which is called synonymous image cloning system. In con- sideration of the large time complexity of the semantic matching process, CUDA parallel programming model and OpenMP multithread programing method of the heterogeneous multicore environment were used to accelerate the process and effectively improve the performance of overall system. Experiments show that this system guarantees higher user involvement, stronger practicability and ensures the users＇ satisfaction of the cloning speed and effect.

关 键 词：图像编辑 场景识别 语义匹配 无缝融合 

分 类 号：TP391.41[自动化与计算机技术—计算机应用技术]