文档下载 文档下载导航
设为首页 | 加入收藏
搜索 请输入内容:  
 导航当前位置: 文档下载 > 所有分类 > 人文社科 > 文学研究 > Comparing and Evaluating Interest Points
侵权投诉

Comparing and Evaluating Interest Points

are presented. In section 5 we select the detector which gives the best results, and show that the quality of its results is very high.

2 Interest point detectors A wide variety of interest point detectors exist in the literature. These detectors can be grouped into three categories. Those in the rst one extract contours and search for maximal curvature or in exion points along these contours (cf. section 2.1.1). Those in the second one extract interest points directly from the greyvalue signal (cf. section 2.1.2), and those in the third one t a template to the signal (cf. section 2.1.3).

2.1 State of the art

2.1.1 Contour based methods The idea of contour based methods is either to search for maximal curvature or in exion poi

nts along the contour chains, or to do some polygonal approximation and to search for particular points such as intersection points. Such methods have existed for a long time we are going to explain some of the more recent ones in detail. Asada and Brady 1] extract interest points for 2D objects from their planar curves. They observe that these curves have special characteristics: the changes in curvature. These changes are classi ed in several categories: corners, endings etc. To achieve robust detection, their algorithm is integrated in a multi-scale framework. A similar approach has been developed by Mokhtarian and Mackworth 16]: they use in exion points of a planar curve. Medioni and Yasumoto 15] use B-splines to approximate the contours. Interest points are maxima of curvature which are computed from the coe cients of these B-splines. Horaud et al. 12] extract line segments form the image contours. These segments are grouped in order to obtain an intermediate representation. Intersections of grouped line segments are the interest points.

2.1.2 Signal based methods Signal based methods compute the measure which indicates the presence of an interest point directly form the signal. Beaudet 2] developed the rst signal based interest point detector. His detector uses the second derivatives of the signal for computing the measure\DET": DET= I I; I 2 where I (x y) is the intensity surface of the image. This measure is invariant to rotation and is related to the Gaussian curvature of the signal. Points where this measure is maximal are interest points. The detector of Moravec 17] is based on the auto-correlation function of the signal. This function measures the di erences between a window of the signal and its four neighbouring windows. If the minimum of these four di erences is superior to a threshold, an interest point is present. However, the neighbourhood is discrete and the directions used are parallel to the rows and columns of the image. xx yy xy

上一页第4页

你可能喜欢

热门文档

相关文档

站点地图 | 文档上传 | 侵权投诉 | 手机版
本站所有资源均来自互联网,本站只负责收集和整理,均不承担任何法律责任,如有侵权等其它行为请联系我们.
文档下载 Copyright 2013 doc.guandang.net All Rights Reserved.  -  email