Abstract—Region segmentation and edge detection are standard image processing operations. Clustering can be used for region segmentation. However, often clustering results depend on the selection of various parameters, such as the number of clusters, or the clustering algorithm used. The framework presented here employs the result of edge detection on the original image, as well as on the clustering results of the same image, to automatically select (according to some agreement measure) the optimal number of clusters, and the corresponding (best) segmentation. The framework supports an extended pixel representation in which other information, such as texture, can be incorporated in addition to edge and region information. To illustrate this framework, the edge guided clustering algorithm presented here, uses the canny edge detection approach to guide region identification through fuzzy k-means clustering. Experimental results on benchmark images for which manual segmentation is available as reference illustrate the effectiveness of this approach.
Index Terms—Image segmentation, fuzzy k-means clustering, canny edge.
S. Visa is assistant professor in the Department of Mathematics and Computer Science at The College of Wooster, Wooster, OH 44691, USA (email: svisa@wooster.edu).
A. L. Ralescu is the director of the Machine Learning and Computational Intelligence Laboratory and professor in the School of Computing Sciences and Informatics at the University of Cincinnati, Cincinnati, OH 45221, USA (email: Anca.Ralescu@uc.edu).
M. Ionescu is computer vision scientist at the Charles River Laboratories (SPC), Seattle, WA 98104, USA (email: Mircea.Ionescu@crl.com).
[PDF]
Cite: Sofia Visa, Anca L. Ralescu, and Mircea Ionescu, "A Framework for Segmentation Using Edge Guided Image Clustering," International Journal of Information and Electronics Engineering vol. 2, no. 1, pp. 29-37, 2011.
