项目名称: 基于鲁棒相似性测度的含噪图像分割的谱聚类方法
项目编号: No.61202153
项目类型: 青年科学基金项目
立项/批准年度: 2013
项目学科: 计算机科学学科
项目作者: 刘汉强
作者单位: 陕西师范大学
项目金额: 25万元
中文摘要: 传统的谱聚类算法应用于图像分割时,相似性测度的构建没有利用图像像素的空间信息,使得算法对于被噪声污染的图像无法获得令人满意的分割性能。本项目利用图像中各种有效的空间信息,分别从像素级、灰度级和区域级三种角度构建融合空间信息的相似性测度,提出相应的融合空间信息的谱聚类算法。此外,还将引入一定量的监督信息与图像的空间信息结合使用,构建更加有效的相似性测度,提出融合空间信息的半监督谱聚类算法。 图像分割是图像处理中的重要环节,分割的质量和速度对后续的图像分析、理解与描述等任务将产生直接的影响。谱聚类是解决图像分割问题的有效方法之一,本项目充分利用图像本身蕴含的空间信息和用户提供的少量监督信息构建相似性测度,旨在解决谱聚类在含噪图像上的分割问题,因此开展本项目的研究具有一定的理论意义和应用价值。
中文关键词: 图像分割;空间信息;区域信息;相似性测度;谱聚类
英文摘要: Due to the construction of the similarity measure not utilizing any spatial information of pixel in the image, traditional spectral clustering algorithms cannot obtain the satisfactory segmentation performance for images corrupted by noise. This project adopts some kinds of effective spatial information derived from the image to design novel spatial information-based similarity measures in terms of pixel level, gray level and region level in the image, and proposes the corresponding spectral clustering algorithms with spatial information. In addition, some supervised information is further incorporated with the spatial information to construct more effective similarity measures for spectral clustering, and then semi-supervised spectral clustering algorithms with spatial information are presented in this project. Image segmentation is one of important aspects in image processing and its quality and speed can have a direct influence on image analysis, understanding and description. Spectral clustering is one of effective methods for image segmentation. In order to solve the problem of spectral clustering applying to noisy images, this project fully utilizes the spatial information embedded in the image and the little supervised information provided by users to design similarity measures for spectral clustering.
英文关键词: image segmentation;spatial information;region information;similarity measure;spectral clustering