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

Figure 1: Interest points detected on the sower painting of\Van Gogh". The detector used is PreciseHarris which is an improved version of the Harris detector. matching as the general context. Image matching based on interest points has given very good results. The quality of the results makes algorithms based on matching reliable. For example 26] obtains excellent results when computing the epipolar geometry from matched interest points. It is also possible to retrieve images from a large collection of images using image matching based on interest points 24]. In the context of image matching, interest points should be geometrically stable under di erent transformations (repeatability). If we want to match two images, at least a subset of the points have to be detected at the same relative position, otherwise no feature-based matching is possible. Repeatability signi es that we obtain the points with the same pre-image independently of changes in the imaging conditions: a 3D point should be detected in both images if it is detected in one of them. This means that the results should be invariant to image transformations. The repeatability rate measures the percentage of points which have been repeated between two images. It has been compared for di erent detectors in the presence of di erent transformations: image rotation, scale change, light changes, viewpoint change and image noise. To illustrate this criterion an example is presented in gure 2. It shows interest points detected on the same scene as in gure 1 under image rotation and scale change. There is an image rotation of 155 between the left image of gure 2 and the image in gure 1. We can see that the repeatability of the interest points detected by PreciseHarris is good: the repeatability rate is 92%. Between the right image of gure 2 and the image in gure 1 the scale change is 1.4. The repeatability rate is 71%. If the greylevel information is used for matching, signal patterns of the points should be distinctive. If all the interest points detected have very similar patterns, it is not possible to distinguish between them: matching based on signal information will fail any point might match with any other point. Information content measures the distinctiveness of the local greylevel patterns at interest points. To measure information content, we describe the local pattern using rotationally invariant combinations of derivatives. The entropy of the distribution of the derivatives for a set of interest points is then measured. 2

Comparing and Evaluating Interest Points

上一页第2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