物体遮挡在二维图像中普遍存在，实质上形成了对计算机而言的“视觉盲点”区域，直接制约计算机自主分析解读图像内容的能力。本项目拟针对该现象展开研究，在大规模信息资源支撑下，设计方法实现对一般性物体遮挡的自动分析补偿。该研究属于涉及多个知识层次的综合性基础应用研究课题，包括遮挡分析判别、补偿信息获取、信息建模、补偿信息融合等多个具体组成环节。本项目将重点针对其中5个核心科学问题展开研究：（1）从物体完整性角度入手，分析遮挡发生的一般性规律；（2）设计关联各内在规律的逻辑架构，整合形成一般性遮挡的统一判定框架；（3）研究基于语义线索从海量数据集中“定点”筛选补偿信息的方法；（4）从物体控制点描述序列的自动分析获取入手，研究原始补偿信息的调整控制和高级建模方法；（5）研究以部分内容匹配为线索驱动补偿信息与原物体图像的全局匹配融合。此外，为搭建完整的补偿方法框架，还将对一系列附属研究内容展开适当研究。

Object occlusion is quite common in two-dimensional images, which actually forms the "vision blind spot" area to computers, and hence directly restricts the ability of computer to analyze and interpret the image content autonomously. The project aims to carry out a research on this phenomenon, and design a method to achieve the automatic compensation of general object occlusion with the support of massive information. The research is a comprehensive fundament and application topic involving many knowledge levels, and is composed of many components such as occlusion analysis and judgment, compensation information retrieval, information modeling, compensation information fusion, and so forth. The project will mainly focus on five core scientific issues within those components: (1) analyze the general rules of occlusion from the perspective of the integrity of object; (2) design a logical architecture that associates internal rules, and then form a unified framework of general occlusion judgement; (3) study the method of “customized” compensation information selection from massive dataset based on semantic cues; (4) study the method of raw compensation information adjustment and advanced modeling, from the view of the automatic retrieval of object landmark sequences; (5) study the global matching and fusion of the compensation information and raw object image, driving by the cues of partial content matching. In addition, in order to build an intact compensation method framework, a series of ancillary contents will be moderately studied as well.