复杂场景视觉注意对象分割方法研究

项目名称： 复杂场景视觉注意对象分割方法研究

项目编号： No.61271289

项目类型： 面上项目

立项/批准年度： 2013

项目学科： 无线电电子学、电信技术

项目作者： 李宏亮

作者单位： 电子科技大学

项目金额： 70万元

中文摘要： 复杂场景分析是智能多媒体信息处理领域中的重要研究课题，而如何建立对象语义模型则是影响场景内容分析的关键因素。本项目针对目前复杂场景语义发现的研究现状，提出了复杂场景的视觉注意对象的表示和分割新方法。由于利用了人的认知心理特性，并把该特性嵌入于高层视觉语义描述中，因此该方法比传统的对象分割方法具有更好的语义特性。通过分析不同视觉特征的显著性表现,建立特征视觉注意度等概念的形式化描述，构建符合视觉认知的局部和全局注意力模型。针对复杂场景对象存在的多样性问题，建立具有视觉注意转移特性的对象描述新方法。并在此基础上，利用空域相关特性，构建潜在注意主题学习模型。建立底层特征到注意对象的多层次表示方法，从而实现复杂场景语义内容的发现与分析。这一研究有望为解决复杂场景内容理解提供新的思路和理论依据。

中文关键词： 图像处理；分割；视觉注意力机制；场景分析；

英文摘要： Complex scene analysis has become an important topic in the field of intelligent multimedia information processing. A key factor in the successful scene content analysis is how to build a semantic model for scene objects. After investigating the current status of semantic discovery in complex scenes, this project proposes a new method to represent and segment visual attention object from complex scenes. Since this method utilizes human cognitive and psychological characteristics that are then embedded in high level semantic descriptions, this method can achieve good semantic description compared with the traditional methods. Four issues will be further studied in this project. The first is to build format descriptions of some concepts such as visual attention degree by analyzing the saliency of visual features. The second aims to propose local and global attention models that tally with the human perception. For the diversity issue in complex scene object recognition, the third task is designed to find a new method to describe visual attention objects based on attention transfer mechanism and build a perceptual topic model. Finally, we perform semantic content discovery and analysis for complex scene by the multi-level description from low-level features to object and event. This work will provide a new idea and

英文关键词： Image processing；segmentation；visual attention；scene analysis；