融合颜色与CSIFT的动画场景匹配技术研究  

Technology of Cartoon Scene Matching Using Color and CSIFT Feature

作　　者：邓莉琼[1] 袁志民[1] 陈丹雯[1] 吴玲达[1,2] 

机构地区：[1]国防科学技术大学信息系统工程重点实验室,长沙410073 [2]装备指挥技术学院重点实验室,北京101416

出　　处：《小型微型计算机系统》2012年第9期2068-2072,共5页Journal of Chinese Computer Systems

基　　金：国家自然科学基金项目(60802080,61002020)资助;国家"八六三"高技术研究发展计划项目(2009AA01Z335)资助

摘　　要：针对产量巨大的动画产业,提出一种综合颜色全局特征和CSIFT(Colored invariant feature transform)局部特征的图像匹配算法,解决了动画场景素材的匹配问题.计算图像H、S、I三个分量的分布情况,生成颜色全局特征;在颜色不变量的基础上,生成CSIFT局部不变特征,并提出两种特征相应的匹配方法,最后将两者从底层的特征提取层和高层的特征决策层按照一定的融合方式进行融合.实验表明颜色特征与CSIFT特征相互弥补,相互利用,极大的改善了动画场景素材的匹配效果,和传统的匹配算法相比,该算法鲁棒性好,在近1000幅的动画素材库中的平均匹配准确度高达80%以上.Aiming at the cartoon industry, an improved matching algorithm which focuses on color feature and CSIPT（ Colored invari- ant feature transform） feature is presented to match the cartoon scenes. By calculating the distribution of three parameters ： H, S, I, the global statistical color feature is obtained. Then the CSIFT feature is calculated based on the color invariance. Meanwhile, two corresponding matching algorithms are brought forward. Finally, the images are matched by fusing the global color and CSIFT fea- tures from two floors： the bottom extraction floor and the top decision floor. Experimental results indicate that the two features make up each other so that it improves matching effect of cartoon scene greatly. Compared with traditional matching algorithms, the algo- rithm is robust; the mean matching precision is up to 80% in the dataset which has near 1000 cartoon images.

关 键 词：颜色 CSIFT算法 动画 匹配 融合 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]