A region-region and region-edge cooperative approach of image segmentation

In the task of segmentation of some complex pictures, it is often difficult to obtain satisfactory results using only one approach of image segmentation. The tendency toward the integration of several techniques seems to be the best solution. The authors introduce a new approach of image segmentation, which tends to combine several sources of knowledge about the image in a way to produce better segmentation results. First, they try to locate germs that are homogenous by means of a region-region cooperative process. A region growing process is then applied to these germs in order to find the region borders. This process is controlled, on the one hand, by the germs' parameters, and on the other hand by a gradient information obtained by a simple edge detector. Finally, a region merging step is applied on the extracted regions in order to reconstruct regions that have been split by the germs' extraction process. This method has given good segmentation results over several complex natural images.<<ETX>>