Directional Edge Tracking for Line Extraction

K.-S. Seo, J.-H. Kim, E.-J. Park, and H.-M. Choi (Korea)


2. Directional Edge Tracking for Line Extraction Image analysis, line extraction, similar orientation, tracking order, fast, orientation distance. In the proposed line extraction, as shown in the Fig. 1, edge image is obtained by the Canny operator, and then line segments are extracted by the proposed directional edge tracking. Thereafter, lines are fitted by the least square fitting algorithm.


In the neural network based methods [8-10], line extractor can be designed by learning process. In those methods, neural networks are learned with various and appropriate learning data, and then the lines can be extracted by the neural network system. Edge tracking based line segment extractor is proposed for fast and accurate extraction of straight lines on a digital image. Edgels (edge elements) are computed by the Canny operator, and then line segments are extracted by the proposed line segment extractor. The extractor is designed on the assumption that every edgel on the same line has similar edge orientation. Connected edgels that have similar orientation are liked and extracted as a line segment. To effectively link and extract line segment, directional edge tracker is proposed. The tracker records the information of tracking direction at the current step, and then at the next step tracking candidates and their priorities are selected by using that information. It is very fast and effective because the number of tracking candidates is always less then conventional 8-neighbor searching method. Finally the extracted line segments are fitted by least square algorithm. The results of experiments show that the proposed method can efficiently extract lines from synthetic and real images. In the similar edge orientation grouping based methods [11-14], connected edgels with similar orientation make a group as a line segment, and then these segments are fitted. LSR (line support region) based method (proposed by Burns et. al. [11]) can effectively detect lines while keeping less time complexity than Hough transform based methods. This method, however, also has relatively high time complexity; so the researches [12-14] to solve this problem have been widely studied. In this paper, directional edge tracking is proposed in which tracking candidates and their priorities are determined by the previous tracking direction in the edge tracking in order to reduce time complexity. Moreover, distance between two edges is computed by integer operation of wraparound mode to reduce processing time.

Important Links:

Go Back