CASIA OpenIR  > 毕业生  > 硕士学位论文
高分辨率DEM图像地形特征线快速提取算法研究
Alternative TitleResearch on High Resolution DEM Image Topographic Feature Line Fast Extraction Algorithm
刘洲俊
Subtype工学硕士
Thesis Advisor胡包钢
2011-05-30
Degree Grantor中国科学院研究生院
Place of Conferral中国科学院自动化研究所
Degree Discipline计算机应用技术
Keyword数字高程模型 山脊线 山谷线 脊线提取 启发式算法 Hessian矩阵 图形处理器 计算统一设备架构 Digital Elevation Model Ridge Valley Ridge Detection Heuristic Algorithm Hessian Matrix Graphics Processing Unit Compute Unified Device Architecture
Abstract图像的曲线结构提取是计算机视觉中一个重要的基本操作,应用十分广泛。在地理信息系统(GIS)和遥感领域中,曲线提取技术通常被用来提取地形结构的特征信息,例如道路,河流,山脊线和山谷线;在医疗影像中常被用来定位和提取各种解剖学特征,例如X光造影中的血管和CT影像中的骨骼。 地理信息系统使用的数字高程模型(DEM)中包含各种地形地貌构造的特征信息,其中,地形特征线(山脊线,山谷线)对于地形研究具有重要的意义。对地形特征线进行自动和高效的提取一直是数字地形分析中的重要内容。目前已经存在一系列的地形特征线提取算法,但是这些算法的局限是计算复杂度较高。另外,随着数据采集设备的发展,数字地形分析中使用的DEM图像的分辨率越来越高,使得现有的地形特征线提取算法效率更低。这就迫切需要一种针对高分辨率DEM数据的高效提取方法。 本文选取从高分辨率DEM灰度图像中山脊/山谷线的快速检测作为研究内容,论文的主要贡献是: 1)提出了一种新的启发式的快速脊线提取算法。在确定山脊线/山谷线上特征点时直接采用固定方向的极值点进行近似处理,之后在特征点连线上采用启发式连线策略。实验结果表明,该算法对于高分辨率DEM图像可以快速提取其中的地形线,算法效率与基于Hessian矩阵的脊线检测算法 [25]相比提高了5至8倍。 2)提出了一种在图形处理器(GPU)上加速基于Hessian矩阵的脊线检测算法 [25]的策略,通过分析原始算法,利用图形处理器上计算统一设备架构(CUDA)的高度并行性来加速算法中计算复杂度高的Hessian矩阵生成模块以及图像特征点提取模块,对于百万像素级的DEM图像该算法可以获得5倍以上的加速比。
Other AbstractExtraction curvilinear structure in images is one of the important basic operations in computer vision, which has been widely used in many other fields. In the fields of geographical information system (GIS) and remote sensing, we extract feature information on terrain structure, such as road, river, ridge and valley; in medical imaging, it is used to locate and extract sorts of anatomy feature, for example, vessels in X-ray imaging and bones in computerized tomography imaging. Digital elevation model (DEM) in GIS includes various terrain feature information, among which curvilinear structures (ridge and valley) play an important role in terrain research. There have been various curvilinear structure extraction algorithms; however the main limitation of them is the high computing cost. Besides, with the development of data collection equipment, the resolution of DEM image used in digital terrain analysis is becoming higher and higher, which makes existing algorithms in a lower efficiency. Based on the above analysis, it is urgent to develop new efficient feature line extraction methods aimed at high resolution DEM images. We take fast ridge/valley detection from high resolution DEM grey image as the research topic. The main two contributions of this thesis are as follows: 1) We propose a new fast heuristic ridge detection algorithm. The algorithm takes the extreme points in certain directions as candidates to determine the feature points in ridges or valleys. Then we use a heuristic strategy to link the feature points to ridges or valleys. Experiments shows that our algorithm is 5-8 times faster than the ridge detection algorithm based on Hessian matrix [25]. 2) We propose an efficient strategy to speed up the ridge detection algorithm based on Hessian matrix [25] using graphics process unit (GPU). By analyzing the original algorithm, we choose to speed up the most computation intensive modules of the algorithm (Hessian matrix generation and feature point detection) based on compute unified device architecture (CUDA). This method can achieve more than five times speedup to the original algorithm on central process unit (CPU) for large scale DEM images with millions of pixels.
shelfnumXWLW1595
Other Identifier200728014628068
Language中文
Document Type学位论文
Identifierhttp://ir.ia.ac.cn/handle/173211/7583
Collection毕业生_硕士学位论文
Recommended Citation
GB/T 7714
刘洲俊. 高分辨率DEM图像地形特征线快速提取算法研究[D]. 中国科学院自动化研究所. 中国科学院研究生院,2011.
Files in This Item:
File Name/Size DocType Version Access License
CASIA_20072801462806(8486KB) 暂不开放CC BY-NC-SAApplication Full Text
Related Services
Recommend this item
Bookmark
Usage statistics
Export to Endnote
Google Scholar
Similar articles in Google Scholar
[刘洲俊]'s Articles
Baidu academic
Similar articles in Baidu academic
[刘洲俊]'s Articles
Bing Scholar
Similar articles in Bing Scholar
[刘洲俊]'s Articles
Terms of Use
No data!
Social Bookmark/Share
All comments (0)
No comment.
 

Items in the repository are protected by copyright, with all rights reserved, unless otherwise indicated.