项目名称: 基于部件结构的图像协同分割方法研究
项目编号: No.61502084
项目类型: 青年科学基金项目
立项/批准年度: 2016
项目学科: 计算机科学学科
项目作者: 孟凡满
作者单位: 电子科技大学
项目金额: 20万元
中文摘要: 图像分割是计算机视觉和多媒体处理领域中的重要研究内容,而如何从图像数据中高效地发现和提取感兴趣对象仍然是一个挑战性的问题。本项目针对目前语义对象分割的研究现状,提出了基于部件结构的协同分割新方法。由于引入了对象的局部区域关联特性,该方法比传统的对象分割方法具有更好的语义特性。通过分析不同部件的特征表现,建立基于深度学习的区域部件描述,构建了初始部件的回归生成模型。针对对象的部件关系,提出了具有空间结构关系的部件描述与图匹配方法。并在此基础上,利用部件结构的一致性,构建了基于部件结构的协同分割优化模型。本项目研究有望为解决图像语义分割提供新的思路和理论依据。
中文关键词: 图像分割;协同分割;部件检测
英文摘要: Image segmentation is an important research in computer vision and multimedia processing. How to efficiently discover and extract interesting object from image data is still a challenging task. By investigating the current status of semantic object segmentation, this project proposes a new part structure based image co-segmentation method. Since the relationships among the local parts are introduced, this method can obtain more semantic results compared with the traditional methods. By analyzing the feature representation of different parts, we study the description of the part based on deep learning, and the design of the initial part generation method. Then, we study the part structure representation and matching based on the spatial relationships among the parts. Based on the part structure representation and matching, we study the design of part structure based co-segmentation model based on the part structure consistency. This work will provide a new idea and theoretical foundation for image semantic segmentation.
英文关键词: Image Segmentation;Co-segmentation;Part detection