一种用于复杂目标感知的视觉注意模型

Visual Attention Model for Complex Target Perception

针对自然场景图像中复杂结构目标的快速定位问题,提出一种新的视觉注意模型。对目标进行学习提取显著性图斑,将图斑的特征信息、异质图斑之间的相对位置关系引入视觉注意过程,采用基于图匹配的图斑搜索策略合并与目标特征相似的异质图斑,从而获得注意焦点。与自底向上的视觉注意模型进行实验对比,结果表明该模型能引入复杂结构目标的特征信息和结构信息,降低无效关注次数,提高视觉注意的效率。

A new visual attention model used for rapid perception of complex targets in natural scene is proposed.In the learning process,the model extracts saliency blobs from a given target’s image.Then during the process of attention on a scene image,it adopts a blob searching and merging strategy based on graph matching to guide visual focus to where it looks like the target.The blob searching and merging strategy uses features of learned heterogeneous blobs and their spatial relative positions,which are all recorded during the previous target learning process.Compared with typical bottom-up visual attention model,experiments show that the new method could efficiently introduce feature and structure information of complex target into the process of attention,reduce useless visual focus shifts,and improve the performance of visual attention.The model could be used to locate complex structural targets in natural scene images