In this paper, we propose a method which combines Canny edge detection and steering kernel regression[5]. As the primary features for extraction by low-level processing techniques, edges are the starting points for many computer vision applications. During the past years, there had been many edge detection algorithms proposed. As an almost standard framework, canny edge detector is widely used in image processing area and often checked by other algorithms for their validity as an almost standard framework. But, due to the deficiency of using direct gradient estimation ( usually the gray-value difference),classical canny method is vulnerable to noise. For overcoming this problem, we combine the steering kernel estimation with the canny edge detection which taking both the spatial and radiometric information into count simultaneously. And the experiment results perform better than classical edge detection method on detail reserving and positioning accuracy under the effect of different kinds of noise.

Date Published: 14 March 2013
PDF: 5 pages
Proc. SPIE 8768, International Conference on Graphic and Image Processing (ICGIP 2012), 87687L (14 March 2013); doi: 10.1117/12.2010363

Published in SPIE Proceedings Vol. 8768:
International Conference on Graphic and Image Processing (ICGIP 2012)
Zeng Zhu, Editor(s)