摘要
针对基于尺度不变特征变换(Scale Invariant Feature Transform,SIFT)的目标识别实时性差的缺陷,提出了一种结合目标主色集(Object Dominant Color Set,ODCS)初定位的SIFT彩色目标快速识别算法(ODCS-SIFT)。将目标识别分为两个阶段:在离线训练阶段,采用人机交互的方式提取目标特征;在实时图像处理阶段,首先基于目标主色集进行顺序网格搜索和种子填充,再根据各主色频数约束来确定目标的初定位区域,最后在灰度化的初定位区域进行SIFT处理。对比实验表明,本方法可有效地提高SIFT目标识别的实时性。
To improve the celerity of object identification based on scale invariant feature transform（SIFT）, a new fast approach called OECS-SIFT was proposed to locate and identify color object from scene using combination of object dominant color set（ODCS） and SIFT. The whole approach comprises two joint stage： off-line training stage and on-line identifying stage. At off-line stage, find OIlS and SIFT libraries through human-machine interaction, while at on-line stage, firstly search in the whole scene image using ordinal grid scanning and seed filling simultaneously, then locate the object roughly using the ODCS frequency restriction, lastly in the smaller grayed location a more real-time and precise SIFT extracting and matching are executed. Experimental results show that the ODCS-SIFT can improve the celerity of object identification effectively.
出处
《计算机科学》
CSCD
北大核心
2009年第12期257-258,266,共3页
Computer Science
基金
国家863高技术研究发展计划资助项目(2006AA04Z212)
河北省教育厅自然科学研究计划项目(Z2008473)资助
关键词
目标主色集
目标识别
特征联合
Object dominant color set, Object identification, Feature combination
