To search, Click below search items. |
| All Published Papers Search Service |
|
|
|
Title |
Detect and Represent Lines Based on a Set of 3-Pixel Elementary Units |
|
Author |
Wei Hui, Wang Zengjin, Liu Bin |
| Citation |
Vol. 6 No. 9 pp. 12-23 |
|
Abstract |
A key task of digital computer image understanding is to organize image pixels according to semantic structure, and one of the most important clues to this task are the edges and profiles of the objects. In most cases edges can be represented as lines, therefore detecting line elements is an important procedure of image understanding. However, edge detection algorithms only provide a set of discrete pixels instead of direct descriptions of lines. In this paper, we propose a new method of line detection and line description based on combinations of multiple three-pixel basic patterns. The new algorithm connects those adjacent pixels that are geometrical likely to form line segments, thus it remarkably simplify the description of an image by using line segments instead of pixels. The proposed method consists of three major components: first, patterns of elementary unit that compose saw-toothed line segment in grid mode are defined. Second, the rule of combination of different basic patterns is provided, and proved to be feasible in experiments. Third, a hierarchical and parallel network computing model to obtain the basic patterns of lines and a clustering algorithm based on basic patterns. Experimental results show that this new method can considerably improve performances in computation time, memory requirement and detection accuracy compared with other classical line detection algorithms. Another advantage of this kind of geometric-character-based clustering algorithm is that the output of this method can be taken as the input of subsequent object for recognition procedures directly. Moreover, no extra manipulations with higher algorithm complexity, such as searching line end points, filtering false lines or segmenting collinear line segments are needed |
|
Keywords |
Line detection, Image understanding |
|
URL |
