融合运动和空间关系特性的显著性区域检测  

Salient region detection of interfusing motion and spatial relationships

作　　者：刘晓辉[1] 金志刚[1] 赵安安[2] 卫津津[1] 

机构地区：[1]天津大学电子信息工程学院,天津300072 [2]天津大学软件学院,天津300072

出　　处：《华中科技大学学报（自然科学版）》2013年第6期45-49,共5页Journal of Huazhong University of Science and Technology(Natural Science Edition)

基　　金：国家高技术研究发展计划资助项目(2009AA01A336)

摘　　要：为了解决传统视觉显著性模型在复杂现实场景中容易失效的问题,提出了一种融合运动和空间关系特性的显著性检测方法.在颜色对比度的基础上,引入图像中目标的空间深度关系和运动特征.首先建立深度显著性模型,并使之与颜色显著性融合获得空域显著图;然后通过分析运动矢量得到运动特征显著性,并采用加性融合和乘性融合联合的方式计算出最终的显著图.实验分析建立在多个立体图像数据库之上,量化评价和视觉观测均表明:本方法可以有效地提取显著性区域,检测效果优于传统方法.Traditional visual salient models are likely to fail in complicated scenes. Thus, a salient de- tection approach to interfusing motion and spatial relationships was proposed. The spatial depth rela- tionships and motion features were added on the basis of color contrast. First, a depth saliency model was built and interfused with color saliency to obtain spatial saliency map. Then the motion feature saliency, acquired by analyzing motion vectors, was applied to the process of computing the final sali- ency map with additive and multiplicative integration. Experiments were analyzed by several stereo image databases. Both quantitative evaluation and direct observation indicate that the approach can de- tect saliency regions effectively and the results are better than that of traditional methods.

关 键 词：视觉显著性模型 深度显著性 运动特征显著性 加性融合 乘性融合 

分 类 号：TP751[自动化与计算机技术—检测技术与自动化装置]