基于多特征与水平集融合的遥感图像分割算法研究

项目名称： 基于多特征与水平集融合的遥感图像分割算法研究

项目编号： No.61502435

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： 吴庆岗

作者单位： 郑州轻工业学院

项目金额： 19万元

中文摘要： 特征提取是图像分割要解决的关键问题之一，该问题的有效解决将有助于提高图像分割的精度。本项目拟结合遥感图像具有多种特征的特点，重点研究其纹理特征和形状特征的提取与表示问题，通过与水平集方法的融合以提高遥感图像分割的精度。首先，项目拟结合视觉感知理论，利用字典学习方法构造纹理基元字典，通过主成分分析和稀疏编码策略研究纹理基元的确定问题，构建目标的纹理特征表示方式。然后，在提取目标特征点的基础上，拟采用以狄利克雷过程为先验信息的贝叶斯模型来描述形状的非参数统计规律，避免对参数的过度依赖，试图提高形状特征提取的鲁棒性。最后，在水平集框架下，研究多特征均值-方差和统计分布两种融合方法，在此基础上拟提出基于多特征和水平集融合的遥感图像分割算法，以提高分割精度，进一步结合分裂算子法来提高求解效率。该项目研究将对遥感图像分割研究起到关键的促进作用，也将为进一步研发具有实用价值的遥感图像处理系统奠定基础。

中文关键词： 字典学习；非参数统计；特征提取；图像分割；水平集

英文摘要： Feature extraction is one of the key problems in image segmentation, which will improve the segmentation accuracy. By means of the properties of multiple features of remote sensing images, we shall focus on the extraction and representation problems for texture and shape of the involved images. The combination of the multiple features with level set method will be explored to improve the segmentation accuracy. Firstly, considering the visual perception theory, the dictionary learning method will be used to construct texton dictionary. The textons will be determined by the strategies of principal component analysis (PCA) and sparse coding (SC). The sparse representation for the texture object of interest will be constructed by the texton dictionary. Then, Dirichlet Process will be adopted to describe the non-parametric statistical rule for the object shape on the basis of feature points so as to improve the robustness of shape feature extraction. This method has the advantage of avoiding the dependence of shape feature on various parameters. Finally, a novel segmentation algorithm will be proposed on the basis of the two fusion strategies, i.e., mean-variance and statistical distribution of various features to improve the segmentation accuracy. Meanwhile, Split-Bregman method will be considered so as to improve the segmentation efficiency. The study will play a key role in promoting the research of object segmentation in remote sensing images, and lays the foundation for the further research and development of the remote sensing image processing system with practical values.

英文关键词： Dictionary learning ;Nonparametric statistics;Feature extraction;Image segmentation;Level set