生物启发的视觉目标搜索和定位研究

项目名称： 生物启发的视觉目标搜索和定位研究

项目编号： No.61272320

项目类型： 面上项目

立项/批准年度： 2013

项目学科： 自动化技术、计算机技术

项目作者： 苗军

作者单位： 中国科学院计算技术研究所

项目金额： 80万元

中文摘要： 生物视觉系统因其良好的注意选择、特征捆绑和高效的语义结构化目标表达机制，使之与目前的人工视觉系统相比，在视觉搜索和目标定位上表现出了显著的优越性。借鉴生物机制、研究生物启发的视觉目标搜索机理与系统，对促进计算机视觉研究及其实际应用有很大意义。本项目主要借鉴生物视觉的"where-what"信息加工通路结构、参照认知领域的拓扑知觉组织理论，从大范围拓扑连通性检测入手，开展拓扑区域的显著性、不变性、表观、形状和物体结构、轮廓特征的提取研究，为实现物体或目标的特征捆绑和语义结构化表达提供有力的支持。本研究拟解决以下关键问题：(1)视觉搜索的快捷路径选择和目标位置预测；（2）物体和目标模型简洁高效的结构化表达和相似性度量；(3)物体和目标模型之间的交互匹配影射和迭代优化计算，以实现目标的精确定位和分割。本研究期待在大范围拓扑性质感知、物体和目标模型的结构化表达这两项关键技术上取得成绩。

中文关键词： 生物启发；注意选择；目标表达；目标搜索；

英文摘要： Biological vision system shows more powerful performance in target search and locating for its perfect mechanisms of attention, feature binding and structural semantic representation of objects, compared with current artificial vision systems. Introducing biological mechanisms in developing the system of biology inspired target search and locating is much meaningful to research of computer vision and applications. This project mainly introduces "where-what" information processing paths. With reference to the perception organizing theory, beginning with the large range topology property detection, we carry out research on detecting saliency, invariance, appearance,shape, structure and contour features of objects in order to support feature binding and structural representation of objects strongly. This research wish to solve the following problems: (1) quick selection of visual search paths and prediction of target positions；(2) concise and efficient structural representation of objects and target models and their similarity measurement；(3) mutual matching-mapping and iterative optimization for accurate target locating and segmentation. This research expects a good harvest of key techniques of large range topology property detection and structural representation of objects and target models.

英文关键词： Biology-inspired；attention；object representation；target search；