Title: A genetic algorithm-based clustering and two-scan labelling for colour image segmentation

Authors: F.Z. Bellala; Feryel Souami

Addresses: LRIA Département d'Informatique, Faculté d'Electronique et d'Informatique, Université des Sciences et de la Technologie Houari Boumediene, Algiers, Algeria ' LRIA Département d'Informatique, Faculté d'Electronique et d'Informatique, Université des Sciences et de la Technologie Houari Boumediene, Algiers, Algeria

Abstract: In this paper, we present a two-step segmentation method. The first consists of colour image quantisation by genetic algorithm-based clustering method. The second consists of connected component labelling (CCL) of quantised image. In the first step, we use real codification of chromosomes and a variable string length to adjust the number of colours of the reduced palette. We use a fitness function with a smallest number of parameters to improve run time. Once pixels are classified in 3D colour space, a two-scan CCL adapted to colour image is proposed and applied to the 2D image plan to get separate regions. To reduce regions number, small regions are grouped with the nearest surrounding regions according to their colour feature. Segmentation results are discussed in RGB and lab colour spaces.

Keywords: clustering algorithms; colour quantisation; connected component labelling; CCL; FCM algorithm; genetic algorithms; image segmentation; colour images; image processing.

DOI: 10.1504/IJCVR.2014.059367

International Journal of Computational Vision and Robotics, 2014 Vol.4 No.1/2, pp.86 - 98

Received: 28 Mar 2013
Accepted: 30 Jul 2013

Published online: 18 Feb 2014 *

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