Title: Colour image segmentation using self-adaptive watershed and affinity propagation clustering

Authors: Qiang Cai; Yaqi Liu; Jian Cao; Haisheng Li

Addresses: School of Computer and Information Engineering, Beijing Technology and Business University, No. 11, Fucheng Road, Haidian District, Beijing 100048, China ' School of Computer and Information Engineering, Beijing Technology and Business University, No. 11, Fucheng Road, Haidian District, Beijing 100048, China ' School of Computer and Information Engineering, Beijing Technology and Business University, No. 11, Fucheng Road, Haidian District, Beijing 100048, China ' School of Computer and Information Engineering, Beijing Technology and Business University, No. 11, Fucheng Road, Haidian District, Beijing 100048, China

Abstract: In this paper, we propose a method for colour image segmentation using self-adaptive watershed and affinity propagation clustering. Firstly, the input image is smoothed by the spectrum envelope filter, the colour gradient is computed on the smoothed image and regional minima are marked using self-adaptive H-minima transformation method. The watershed transform is used to segment the marked gradient image we get in the previous step. Then, affinity propagation clustering is applied to optimise the segmentation using colour moments computed on each local region. Since colour gradient and colour moments are used, the proposed algorithm makes the best of colour information. Experimental results show that the proposed algorithm is more suitable for colour image segmentation, can achieve good segmentation performance and overcomes the over-segmentation problem in watershed.

Keywords: colour images; image segmentation; self-adaptive watershed; affinity propagation clustering; colour information; colour gradient; colour moments.

DOI: 10.1504/IJCAT.2016.076774

International Journal of Computer Applications in Technology, 2016 Vol.53 No.4, pp.315 - 322

Received: 15 Feb 2014
Accepted: 11 Aug 2014

Published online: 01 Jun 2016 *

Full-text access for editors Full-text access for subscribers Purchase this article Comment on this article