A feature-based region growing-merging approach to color image segmentation

Color image segmentation, a problem with more than one solution, could be faced as a process of categorizing a color image into several homogen regions containing similar objects. In this paper a new and effective unsupervised color image segmentation method is introduced which utilizes three main kinds of features. These features fall in the domain of color, spatial and texture information. The method tries to treat pixels as particles and provides them with a search space, motivated with Particle Swarm Optimization (PSO) with random motion properties to have better and more effective region growing and merging compared to other search spaces. For the first time pixels have the ability to “move” and “find” other homogeneous pixels or regions. The experiments show promising results compared to existing methods.