CASIA OpenIR  > 毕业生  > 硕士学位论文
宽基线立体图像的区域匹配及应用
Alternative TitleRegion based Wide Baseline Stereo Matching and its Applications
宋小丽
Subtype工学硕士
Thesis Advisor胡占义
2007-05-26
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模式识别与智能系统
Keyword宽基线立体视觉匹配 仿射不变区域 最稳定极值区域 遥感图像处理 建筑物变化检测 自动变化检测 Wide Baseline Stereo Matching Affine Invariant Region Maximally Stable Extremal Regions Remote Sensing Image Processing Building Change Detection Automatic Change Detection
Abstract宽基线立体图像匹配是计算机视觉领域的一个重要研究方向,在三维重建、目标识别以及遥感图像处理等领域都有着重要的研究和应用价值。到目前为止,尽管国内外的科研人员在该方向上取得了丰硕的研究成果,但仍有许多关键问题未能很好地解决。本文围绕宽基线立体图像的区域匹配算法展开了深入地研究和分析,并尝试将其应用到高分辨率遥感图像的变化检测中,主要工作有以下三个方面:实现了一个基于仿射不变特征区域的宽基线立体图像的匹配算法。该匹配算法主要由三个部分组成:(1)特征提取。我们选择了性能优异的仿射不变特征区域——最稳定极值区域MSER(Maximally Stable Extremal Regions);(2)特征描述。使用了SIFT(Scale Invariant Feature Transform)描述子来计算特征区域的描述向量,其中每个MSER对应一个SIFT描述子向量;(3)MSER匹配。根据SIFT描述子向量对MSER进行匹配,并结合图像之间的极线约束去除错误的匹配区域。同时,所实现的匹配算法通过大量具有代表性的图像进行了测试。实验结果表明本文实现的基于MSER的图像匹配算法,在有较大的视角变化、明显的光照变化、以及较大尺度变化等情况下,仍然能够比较准确地找到两幅图像之间的匹配区域,具有较高的可靠性和鲁棒性。 提出了一种基于图像特征匹配的遥感图像建筑物变化检测方法。建筑物变化检测是城市变化检测中的一个重要问题。该方法可以检测两幅高空间分辨率遥感图像中的目标建筑物是否发生了变化,并在无变化时确定该目标建筑物在参考遥感图像上的具体位置。图像特征匹配方法的使用较好地克服了由成像条件不同而产生的图像变化对建筑物变化检测造成的干扰,可以比较好地适用于视角和光照不同以及有局部遮挡存在的情况。我们使用Ikonos、Quickbird卫星遥感图像和航空遥感图像进行了初步实验,实验结果表明该方法可以比较可靠地检测到建筑物变化。 提出了一种使用两幅高分辨率遥感图像进行城市场景变化检测的方法。该方法的基本思想是充分利用遥感图像的成像特点和MSER的性质,并结合图像的特征匹配以及两幅图像之间的极线约束关系来检测城市场景中比较显著的真实变化。对大量卫星遥感图像在不同形变、不同光照条件和不同分辨率下进行的变化检测实验表明:(1) MSER能够有效地表达遥感图像中有意义的场景变化;(2)在视频图像匹配中获得成功的基于MSER特征的匹配方法,应用于遥感影像中同样具有较高的辨析力,能够可靠地辨析场景中因视角不同等原因引起的图像变化,为变化检测过程中有效去除这些图像变化提供了可靠的依据。本文可望为实现遥感图像的自动变化检测提供一种可行的途径。
Other AbstractWide baseline stereo matching is one of the essential problems in computer vision and finds many applications such as 3D reconstruction, object recognition and remote sensing image processing. Although a variety of works have been done in the field, wide baseline stereo matching is still a challenging problem, and many difficult problems persist. This thesis is focused on some practical issues on wide baseline stereo matching, and the main work is summarized as follows:  A MSER (Maximally Stable Extremal Regions) based wide baseline stereo matching algorithm is implemented. It is composed of the following three steps: (1) Feature detection. MSER is used as the feature. (2) Feature description. Each MSER is described by SIFT (Scale Invariant Feature Transform) descriptor. (3) MSER matching. Corresponding MSERs are obtained via SIFT descriptor. The experimental results with typical images show that algorithm can satisfactorily find correspondences across two images and perform well under viewpoint change, illumination variation and large scale change.  A method based on image feature matching is proposed for detecting changed buildings in high spatial resolution remote sensing images. The method examines the possible correspondences of the MSER from the current image to the reference image, and by which to infer whether a specified building has changed. If unchanged, computes its location in the reference image. The use of the feature matching for building change detection can well overcome the problems caused by significant changes such as viewpoint and illumination changes, as well as occlusions. The preliminary experimental results with Ikonos satellite images,Quickbird satellite images as well as aerial images show that the proposed method can satisfactorily detect building changes.  A novel approach for detecting urban changes from a pair of high spatial resolution remote sensing images is explored. The basic idea of the approach is that: MSERs can be used to represent urban change contents. MSERs changes can be considered as the urban change under this basic principle, and the change detection is converted to a MSER matching problem. Varying experiments with satellite images under various conditions such as geometric distortion, illumination variation and resolution difference show that our explored change detection approach performs well, and could be used as a feasible way to solve the problems in remote sensing applications.
shelfnumXWLW1151
Other Identifier200428014628045
Language中文
Document Type学位论文
Identifierhttp://ir.ia.ac.cn/handle/173211/7423
Collection毕业生_硕士学位论文
Recommended Citation
GB/T 7714
宋小丽. 宽基线立体图像的区域匹配及应用[D]. 中国科学院自动化研究所. 中国科学院研究生院,2007.
Files in This Item:
File Name/Size DocType Version Access License
CASIA_20042801462804(4448KB) 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
[宋小丽]'s Articles
Baidu academic
Similar articles in Baidu academic
[宋小丽]'s Articles
Bing Scholar
Similar articles in Bing Scholar
[宋小丽]'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.