项目名称: 基于目标区域演化分割的图像描述研究
项目编号: No.60875010
项目类型: 面上项目
立项/批准年度: 2009
项目学科: 轻工业、手工业
项目作者: 任明武
作者单位: 南京理工大学
项目金额: 28万元
中文摘要: 本项目提出了图像分割的结果多样性并进行描述的观点。图像分割结果的多样性是指对于同一个目标区域随分割参数取值的不同往往会得到不同的结果,若某种图像分割方法能得到目标区域的众多表象并有效地组织和描述,则不同的应用应能从该描述中找到其需要的分割结果。项目研究从图像分割结果多样性的角度出发,研究了基于微分曲率的PDE图像去噪,研究并实现了基于关系稳定性的图像分割方法,研究并实现了用形态学分水岭指导间隙区域归并,并实验证明了其优越性;实现了一种基于显著性的活动轮廓模型,研究了自然纹理目标的灰度不均特性和区域增长的分割方法。在算法方面,研究并实现了一种基于链码的快速分水岭变换,并采用了树结构来描述目标表象演化关系。在本项目的支持和促进下,课题组成员很好地开展了此图像分割及相关图像领域的研究工作,并为后续研究打下了较好的基础。
中文关键词: 图像分割; 图像描述; 金字塔分割; 分水岭分割
英文摘要: This research considered the image description method aimed to diversity of segmentati-on results of one same image. The diversity of segmentation results referred to that many different region results can be got for same image and same segmentation method but with different values of parameter,to find one image description method which can got and descript the diversity of segmentation is very important to applications. The project start from the deversity of image segmentation results,we investigate PDE image denosing based on differential curvature,investigate and execute a image segment method based on relationship stability,investigate and execute a method of using morphological watershed to guide the mergence of the gap region and the experiment proves its superiority;execute an active contour model based on significant,investigate intensity inhomogeneity and natural texture segmentation and segmentation using region growing strategy.In the research of algorithm,we investigate and execute a fast watershed transformation based on chain code and use the tree structure to describe the relations and evolvement between region status.Task group members work well on this image segmentation and related research of image field by the support and promote of the project,and laid a good foundation for the follow-up study.
英文关键词: Image Segmentation; Image Description; Pyramid Structure; Watershed Segmentaion