基于语义联想和视觉焦点的场景目标识别  

Scene target recognition based on semantic association and visual focus

作　　者：降小龙 周海英[1] 

机构地区：[1]中北大学计算机与控制工程学院,山西太原030051

出　　处：《计算机工程与设计》2016年第4期1017-1020,1036,共5页Computer Engineering and Design

基　　金：山西省自然科学基金项目(2013011017-6)

摘　　要：为改进传统识别方法中各种分类器对目标区域分类错误的状况,在分类器中加入语义联想的机制,通过语义网的语法结构对分类出错的区域进行重新分类,直到将目标区域分类满足语法结构;为实现多目标的准确定位,引入视觉焦点吸引力的概念,通过分块操作提取最佳吸引力点来定位目标。实验结果表明,这两个改进将目标识别率提高了2%-8%,实现了对复杂场景中遮挡的多目标数目的确定和具体位置的定位。To improve the error classification situation of a variety of classifiers in conventional recognition methods,a semantic association mechanism was presented to join in the classifiers,and the areas of error classification were reclassified based on the grammatical structure of the semantic Web,until the target area was classified to meet the grammatical structure.To achieve accurate positioning of multiple objectives,the concept of visual focus attraction was proposed.Blocking operation was used to extract the best attractive point to locate the target.Experimental results show that,the two improvements not only increase target recognition rate by 2%to 8%,but also determine the number of occluded multi-targets and position the specific locations in complex scenes.

关 键 词：语义联想 视觉焦点 复杂场景 目标检测 目标识别 

分 类 号：TP391.41[自动化与计算机技术—计算机应用技术]