中文摘要：

图像场景分析是目前计算机视觉领域的研究热点，体现了场景与目标之间的包含关系。在分析过程中合理的使用基于上下文关系的知识可以提高场景分析模型的适用性和目标识别的准确率。从“图像集——场景——目标——部分——视觉词汇”这种层次的角度进行场景分析，将全局上下文信息和局部上下文信息同时融入到基于HDP的生成图模型中，在场景层和目标层这两个不同的层次上，共同作用于场景分析。场景分析的结果可以用来约束目标识别，目标识别的结果可以反馈作用于场景分析。

英文摘要：

Currently, image analysis is the central issue in the field of computer vision. It reflects the inclusion relationships between the scene and the object. In the process of scene analysis, properly used contextual knowledge can improve the applicability of the scene analysis model and the accuracy of the object recognition. We do our research on the scene a- nalysis from a hierarchical perspective of ＂database-scene-object-part-visual words＂, and we add global contextual informa- tion and local contextual information into a generative graph model based on a hierarchical dirichlet process, analyzing the scene at the scene-level and the object-level at the same time. In this way, the result of the scene analysis can be used to constrain object recognition, while the object recognition result feedback effect on the scene analysis.