Abstract:

A color image segmentation algorithm was proposed under the framework of seeded region growing (SRG).Compared to existing SRG methods, this method was robust to the selection of initial seedpixels and the order of region growing. Specifically, there were three main steps: 1) computing a neighborhood based similarity factor (NSF) based on the local color histogram of each pixel; 2) building the criteria of seed selection, region growth and region termination principled by NSF; 3) assigning labels to those un-segmented pixels for a final segmentation. Experimental results showed the superior performance of our methods in computational complexity and segmentation accuracy to the popular method of JSEG.

Key words:  image segmentation; region growing; neighbor similarity factor