SIA OpenIR  > 光电信息技术研究室
前视线外与可见光图像匹配方法研究
其他题名Research on Forward-looking Infrared and Visible Images Matching Method
艾锐1,2
导师史泽林
分类号TP391.41
关键词图像匹配 红外图像 直线段提取 直线编组 隐直线度
索取号TP391.41/A19/2013
页数92页
学位专业模式识别与智能系统
学位名称博士
2013-06-01
学位授予单位中国科学院沈阳自动化研究所
学位授予地点沈阳
作者部门光电信息技术研究室
摘要随着红外成像技术在军事、遥感、医学等领域的广泛应用,红外与可见光图像的匹配问题引起了国内外研究者的普遍关注和深入研究。图像匹配的目的是找出可能在不同成像条件下获取的两幅或多幅图像的共同区域并确定它们之间的几何对应关系。图像匹配可以分为同源图像匹配和异源图像匹配,红外与可见光图像匹配属于异源图像匹配中的一类,其难点在于图像间同时存在着几何差异和光谱差异。红外和可见光图像间较为稳定的是边缘轮廓信息,本文着重研究了直线段特征的提取、扩展及其在匹配中的运用,主要包括:红外成像系统在远距离对地探测时受传输影响导致接感知的场景温差较小,通常表现为图像信噪比较低,已有的直线段检测算法难以有效检测出低信噪比图像中的直线段特征。本文定义了区域非各向同性度,数学上证明了直线区域梯度具有相位一致性;通过统计计算图像原始梯度场上每个图像点邻域内梯度分布,提出了一种新的稳健的直线段检测算法。对比实验表明本文的直线段检测算法更能有效的抑制噪声干扰。针对异源图像灰度信息不可用作匹配,而单条直线段所包含的信息难以进行可靠匹配的实际问题,本文将局部邻域中的直线段组合成具有仿射不变性的直线编组,得到了一种稳定的中层特征,给出了特征描述及图像匹配应用方法。实验表明该算法能有效的实现宽基线条件下红外与可见光的匹配问题。窄基线条件下异源图像匹配可简化为参数优化问题。研究发现,基于隐相似度的匹配算法中函数最优值与正确的空间映射关系不对应。针对该问题,本文对直线特征的内涵及其匹配处理进行了泛化,提出了一种新颖的称为隐直线度的相似度量作为目标函数,等效地将基准图直线段特征和实时图的图像区域进行匹配,进一步通过多分辨率分析,由粗到精用Powell算法进行参数寻优得到了正确的空间映射关系。实验结果表明,本文算法能够有效实现红外与可见光图像的自动配准,并且在实时性上比已有算法有很大提高。最后指出自动目标识别难点是类目标识别,已有的基于图像外观灰度的特征提取算法并不适合于类目标识别问题,边缘轮廓形状这些结构特征才是可靠的类别特征。由于巨大的目标类别数据库使得为每个类别都建立模型然后依次计算的处理方式难以满足实时性要求,需要更加有效的中层形状特征提取方法以提供类别目标的快速索引。图像的结构化特征提取和中层形状特征编组是重要的研究方向。
其他摘要With the infrared imaging technology widely used in military, remote sensing, medical and other fields, the problem of the infrared and visible images matching has been the focus of attention in recent years at home and abroad. The purpose of images matching is to find out to obtain two or more common area in the images under different imaging conditions and then to determine the geometry corresponding relationship between them. Image matching can be divided into same and different source image matching. Infrared and visible images matching belong to different source images matching. And the difficulties in different source images matching are image geometry differences and spectral differences. At the same time, the most stability between infrared and visible images is edge contour information. So this dissertation studies the characteristics of straight line segments extraction, extension, and its use in the images matching. Because of the infrared imaging system used in long-range detection is seriously affected by the transmission, the temperature difference of the scene is small and signal to noise ratio of image is very low. In this case, all the existing line segment detection algorithms are difficult to effectively detect the straight line features in the low SNR images. This paper defines the regional non-isotropic degrees. The linear area gradient phase consistency were justified by mathematics. Through getting each image point neighborhood gradient distribution for the whole original image gradient field, this paper proposes a new steady line segment detection algorithm. Comparative experiments show that this line segment detection algorithm can effectively suppress noise interference. According to the different source image gray-level information cannot be used as a matching and the information in a single straight segment cannot used in the practical problems, this article make line segments in local neighborhood into affine invariance group. Then a stable middle features will be got and character description of image matching method is presented. Experiments show that the algorithm can effectively realize matching problem of infrared and visible light in the wide baseline condition. Under the narrow baseline condition, different source images matching can be simplified into parameter optimization problems. Studies have found that the optimum value of the space mapping is not correct in the matching algorithm based on implicit similarity function Aiming at this problem, the connotation and characteristic of straight line matching processing has carried on the generalization in this paper. We proposed a novel called implicit straightness of similar measure as the objective function. It is equivalent to a benchmark figure line segment features and real-time image area of the image matching and further through the multi-resolution analysis, from coarse to fine with Powell algorithm for parameter optimization to get the correct space mapping relation. The experimental results show that the algorithm can effectively realize the infrared and visible light image automatic registration and has a lot improvement in real-time performance than the existing algorithms. Finally we points out that the difficulties in the automatic target recognition are similar target recognition. The existing feature extraction algorithm based on image gray appearance is not suitable for target recognition problem and the edge contour shape category characteristics of these structures are reliable. Target classes due to huge database make the model difficult to meet the real-time requirements. That needs more effective middle shape feature extraction method to provide category target fast index. Structural characteristics of the image shape feature extraction and middle marshaling is an important research direction.
语种中文
产权排序1
文献类型学位论文
条目标识符http://ir.sia.cn/handle/173321/10767
专题光电信息技术研究室
作者单位1.中国科学院沈阳自动化研究所
2.中国科学院大学
推荐引用方式
GB/T 7714
艾锐. 前视线外与可见光图像匹配方法研究[D]. 沈阳. 中国科学院沈阳自动化研究所,2013.
条目包含的文件
文件名称/大小 文献类型 版本类型 开放类型 使用许可
前视线外与可见光图像匹配方法研究.pdf(4307KB) 开放获取CC BY-NC-SA请求全文
个性服务
推荐该条目
保存到收藏夹
查看访问统计
导出为Endnote文件
谷歌学术
谷歌学术中相似的文章
[艾锐]的文章
百度学术
百度学术中相似的文章
[艾锐]的文章
必应学术
必应学术中相似的文章
[艾锐]的文章
相关权益政策
暂无数据
收藏/分享
所有评论 (0)
暂无评论
 

除非特别说明,本系统中所有内容都受版权保护,并保留所有权利。