CORC  > 自动化研究所  > 中国科学院自动化研究所  > 毕业生  > 博士学位论文
题名图像特征检测及应用
作者戴志军
学位类别工学博士
答辩日期2009-05-28
授予单位中国科学院研究生院
授予地点中国科学院自动化研究所
导师吴毅红
关键词特征检测 仿射不变 边缘检测 聚焦区域 角点检测 图像质量评价 特征匹配 feature detection affine invariant edge detection focused region corner detection image quality assessment feature matching
其他题名Image feature detection and its applications
学位专业模式识别与智能系统
中文摘要本文针对图像特征检测及应用进行了深入的研究,涉及到了图像处理的一些基本问题,包括区域特征检测、边缘特征检测和角点检测等。图像特征检测是计算机视觉中的一个关键问题,在物体识别、三维重建、图像配准、视频理解、图像检索等诸多领域具有非常重要的应用。尽管近些年来对该问题的研究取得了一些突破性进展,但由于图像成像模型多样、成像条件复杂、场景多样和存在各种几何形变,图像特征检测及应用依旧是一个极具挑战性的热点研究课题。本文针对这一问题进行了深入研究,并给出了相应的应用。论文工作的主要创新有以下三个方面: 1. 在区域检测方面, a). 提出了检测图像聚焦区域的方法,并应用到图像质量评测和图像检索上。设计了一个聚焦程度检测器,通过从图片中生成聚焦程度映射图来实现。聚焦程度检测的结果能用来获得图片的聚焦区域。关于聚焦程度检测的应用,以图像检索和图像质量评测为例进行了探索。基于聚焦程度检测器和映射图的检索算法能够得到更准确的检索结果,所提出的图像质量评价算法在分辨高质量图像和低质量图像上有很强的鉴别力。 b). 提出了一种新的图像区域特征检测方法并应用到图像匹配上,该方法的主要思想是从高斯差分空间寻找最大稳定极值区域。检测方法拟合出的椭圆区域可作为仿射不变区域检测器。在图像的整个高斯差分空间可以检测到大量的区域。检测的结果在图像匹配上有着非常卓越的性能。在保证正确匹配率不低的情况下,该方法所得到的匹配数要远远超出以前提出的仿射不变区域检测方法。 2. 在边缘检测方面,提出了一种基于构造组件树的边缘检测新框架,并应用到稳定边缘提取和曲线匹配上。这个框架在边缘属性计算上非常高效并且非常方便后续的图像处理。 此外,这是一个对用户完全开放的框架。通过定制框架准则,可以根据目标边缘的属性检测到所需要的边缘。基于该框架的边缘检测算法显示了对阀值的低依赖性。本方法提取大量稳定的边缘在曲线匹配上可以获得更多的正确匹配。 在角点检测方面,提出一种自动对应标定板上角点的方法,基于此提出准确计算图像畸变中心的方法。根据相机和场景的不同情形,本文给出不同的标定畸变中心的方法。估计的结果可应用到图像校正和三维重建。使用本文的方法所计算出的畸变中心可使图像校正和三维重建的精度得到提高。方法主要是基于几何不变量,这样不仅准确而且非常便捷,实施过程中不需要单应和极几何估计,对使用径向相机的视觉系统很有帮助。
英文摘要In this paper, we investigate the image feature detection and its applications. This field contains many essential image processing problems, such as region feature detection, edge feature detection and corner detection. Image feature detection is a key problem in many applications of computer vision, such as object detection, 3D reconstruction, image registration and video understanding. Though great progress has been made in this field recently, it is still a challenging problem, due to complexed imaging conditions, various scenes, and different shape distortions. In this dissertation, the main work is focused on image feature detection and then gives the corresponding applications. The main contributions are summarized as follows: 1. On region detection, a). we propose a method to detect focused regions in an image and then apply this to image quality assessment and image retrieval. We design a focusing degree detector by which a focusing degree map is generated for a photograph. The results could be used to obtain focused places of photographs. As a concrete example of their applications, image retrieval and image quality assessment are investigated. The retrieval algorithm based on this detector and map can get more accurate retrieval results and the proposed assessment algorithm has a high ability to discriminate photos from low quality to high quality. b). we propose a new image region detection method and then apply this to image matching. The main concept of our method is detecting maximally stable extremal regions under the difference of Gaussian scale space. After each of these regions is fitted with ellipse, the method becomes an affine invariant region detector. We can detect a large amount of such regions in the whole difference of Gaussian scale space. The detected regions have excellent performance on image matching. Not only the matching correct rate is not low, but also the matching count of our method outperforms all previous affine region detectors. 2. On edge detection, we propose a new edge detection framework based on component tree data structure and then apply this to image curve matching. This framework is efficient for edge property computation and convenient for subsequent image processing. In addition, it is open to users. Through customizing framework rules, we detect expected edges according to the targeted edge properties. The edge detection algorithm based on this framework shows the low dependence on thres...
语种中文
其他标识符200618014628038
内容类型学位论文
源URL[http://ir.ia.ac.cn/handle/173211/6175]  
专题毕业生_博士学位论文
推荐引用方式
GB/T 7714
戴志军. 图像特征检测及应用[D]. 中国科学院自动化研究所. 中国科学院研究生院. 2009.
个性服务
查看访问统计
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。


©版权所有 ©2017 CSpace - Powered by CSpace