多变量形态学分水岭理论及其在多通道图像处理中的应用研究

项目名称： 多变量形态学分水岭理论及其在多通道图像处理中的应用研究

项目编号： No.61461025

项目类型： 地区科学基金项目

立项/批准年度： 2015

项目学科： 无线电电子学、电信技术

项目作者： 雷涛

作者单位： 兰州交通大学

项目金额： 43万元

中文摘要： 形态学分水岭理论已广泛应用于单通道图像处理，而拓展形态学分水岭理论到多通道图像(彩色图像、高光谱图像)处理并建立多变量形态学分水岭理论模型是当前的一个热点和难点问题。本项目以经典形态学分水岭理论为基础，围绕构建多变量形态学分水岭理论模型、拓展多变量形态学分水岭理论在多通道图像处理中的应用问题开展深入研究。重点研究基于深度函数结合随机投影理论的多变量极值估计算法；研究多变量形态学算子的对偶性及自对偶特性；探明结构元素尺寸与多变量梯度幅值的函数关系，构建基于空间变结构元素的多变量梯度重建算子；通过理论分析和实验，建立基于形态学扩展最小变换(H-minima)结合最小生成树的多变量局部极值优化模型；探明谱-空间信息的约束关系，最终建立多变量分水岭理论模型，为完善形态学基础理论并实现多通道图像精确分割、高光谱遥感影像地物准确分类及目标识别提供理论和技术支撑。

中文关键词： 图像分割；边缘分割；区域分割；边缘检测

英文摘要： The theory of morphological watershed has been widely used in single channel image processing. It is a hot and difficult topic to extend the applications of the theory of morphological watershed to multi-channel images(color and hyperspectral images) and construct the theoretical model of multivariate morphological watershed. Based on the classical morphological watershed theory, this project aims to do a series of in-depth researches on the construction of theoretical model of multivariate morphological watershed and extension of the theoretical applications in multi-channel image processing. These researches include: (1) to study on multivariate extremum estimation algorithms based on depth function and random projection, (2) to study on the duality and self-duality of multivariate morphological operators, (3) to explore the relationship between the size of structure elements and the multivariate gradient amplitude, (4) to construct the gradient reconstruction operators based on spatial variational structure elements, (5) through theoretical analysis and experiments, to construct multivariate local extremum optical model which is based on morphological extensional H-minima and minimal spanning tree, (6) to explore the constrain relationship between spectral and spatial information, and finally construct a theoretical model of multivariate morphological watershed which will provide theoretical and technological support for perfecting(refining) the basic theory of mathematical morphology and realizing the accurate segmentation of multi-channel images, and the accurate classification of hyperspectral remote sensing images and object recognition.

英文关键词： image segmentation;edge segmentation;area segmentation;edge detection