基于三维分割的太阳光球亮点特征提取技术研究

项目名称： 基于三维分割的太阳光球亮点特征提取技术研究

项目编号： No.11303011

项目类型： 青年科学基金项目

立项/批准年度： 2014

项目学科： 数理科学和化学

项目作者： 杨云飞

作者单位： 昆明理工大学

项目金额： 28万元

中文摘要： 本项目研究太阳高分辨率图像中光球亮点（igBPs）的识别、跟踪和特征提取等图像处理问题。提出了在四维坐标系下进行三维立体分割的思想，旨在提高igBPs处理算法的可靠性和准确性。目前国内外对igBPs的处理基本上采用了"先识别后跟踪"的方式，即先在单幅图像上进行目标分割识别、然后再在时间序列图像中跟踪其演化过程，把识别和跟踪分为两个独立的处理过程。本项目的创新之处在于将序列图像构建成一个包含平面坐标x,y、强度i和时间t的四维坐标系，充分利用时间轴上的关联性，将igBPs的识别与跟踪问题转换成在这个四维坐标系下根据强度进行三维分割和识别其立体结构的问题。研究内容包括三维预处理、igBPs立体结构的分割、筛选和确认，以及发生合并与分裂的复杂igBPs结构的处理。关键技术包括基于三维形态学的不对称滤波和平滑、区域生长算法的种子管理和生长准则定义、不同特征体的区域特性比对和igBPs的精确提取等。

中文关键词： 光球亮点；三维立体分割；边识别边跟踪；目标识别与跟踪；特征提取

英文摘要： Inter-granular Bright Points (igBPs) are small-scale objects in the solar photosphere which can be seen within dark inter-granular lanes. The aim of this proposal is to improve the reliability and accuracy of the algorithm which is used to identify, track and extract igBPs. A four-dimensional coordinate system(x, y, i, t) including the location(x, y), intensity(i) and time(t) is defined to reveal the three-dimensional structure of igBPs. In this coordinate system, both the intensity in a single image and the evolution on the time-dimension are fully exploited, and then a three-dimensional segmentation is employed to identify the structure of igBPs. The evolution of igBPs is reflected on the time-dimension of the three-dimensional structure. This approach differs from the traditional "Detect before Track" method that the igBPs are first detected in each image, and then tracked in time-series images. The studying is focused on the following fields: three-dimensional image preprocessing which consists of smoothing, sharpening, especial preprocessing noises on the time-dimension; obtaining the preliminary structure of igBPs by using the three-dimensional segmentation; identifying igBPs more accurately by filtering out all other structures except igBPs; extracting the complex structures of igBPs that undergo splittin

英文关键词： Inter-granular Bright Points(igBPs)；Three-dimensional segmentation；Track-while-detect；Target detection and tracking；Feature extraction