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 > 高等教育 > 理学 > A Survey of Shape Feature Extraction Techniques
侵权投诉

A Survey of Shape Feature Extraction Techniques

A Survey of Shape Feature Extraction Techniques

A Survey of Shape Feature Extraction Techniques 59

4.3 Discussions

Polygonal approximation technique can be used as a simple method for contour representation and description. The polygon approximation have some interesting properties:

it leads to simplification of shape complexity with no blurring effects.

it leads to noise elimination.

although irrelevant features vanish after polygonal approximation, there is no

dislocation of relevant features.

the remaining vertices on a contour do not change their positions after polygonal

approximation.

Polygonal approximation technique can also be used as preprocessing method for further extracting features from a shape.

5. Spatial interrelation feature

Spatial interrelation feature describes the region or the contour of a shape by the relation of their pixels or curves. In general, the representation is done by using their geometric features: length, curvature, relative orientation and location, area, distance and so on.

5.1 Adaptive grid resolution

The adaptive grid resolution (AGR) was proposed by [24]. In the AGR, a square grid that is just big enough to cover the entire shape is overlaid on a shape. A resolution of the grid cells varies from one portion to another according to the content of the portion of the shape. On the borders or the detail portion on the shape, the higher resolution, i.e. the smaller grid cells, are applied; on the other hand, in the coarse regions of the shape, lower resolution, i.e. the bigger grid cells, are applied.

To guarantee rotation invariance, it is necessary to convert an arbitrarily oriented shape into a unique common orientation. First, find the major axis of the shape. The major axis is the straight line segment joining the two points P1 and P2 on the boundary farthest away from each other. Then we rotate the shape so that its major axis is parallel to the x-axis. This orientation is still not unique as there are two possibilities: P1 can be on the left or on the right. This problem is solved by computing the centroid of the polygon and making sure that the centroid is below the major axis, thus guaranteeing a unique orientation.

Let us now consider scale and translation invariance. We define the bounding rectangle (BR) of a shape as the rectangle with sides parallel to the x and y axes just large enough to cover the entire shape (after rotation). Note that the width of the BR is equal to the length of the major axis. To achieve scale invariance, we proportionally scale all shapes so that their BRs have the same fixed width (pixels).

The method of computation of the AGR representation of a shape applies quad-tree decomposition on the bitmap representation of the shape. The decomposition is based on successive subdivision of the bitmap into four equal-size quadrants. If a bitmap-quadrant does not consist entirely of part of shape, it is recursively subdivided into smaller and smaller quadrants until we reach bitmap-quadrants, i.e., termination condition of the recursion is that the predefined resolution is reached. Figure 15(a) is an example of AGR.

上一页第17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