Colour Image Segmentation based on Multitolerance Region Growing

I. Fondón, C. Serrano, and B. Acha (Spain)


Image segmentation, region growing, L* u* v* colour space, image manipulation and compression


In this paper we propose a new method to segment colour images. It is based on a novel region growing technique whose tolerance step is automatically adapted to make the method useful and totally valid for all kind of images. The selection of the seeds is done with the help of an image of colour distances in the L* u* v* colour space, selecting the local maxima of this image and thresholding them. The step of the region growing is variable in order to obtain the best final result. It begins with a very restrictive value and keeps on growing until the best region is selected. We use a contrast parameter to decide whether a region is the best or not, choosing the one with the highest contrast in relation to the background. This method has been tested on 174 colour images with excellent results in all of them and we give experimental evidence of this effectiveness in this paper.

Important Links:

Go Back