Fast and reliable color region merging inspired by decision tree pruning

In this paper, we exploit some previous theoretical results about decision tree pruning to derive a color segmentation algorithm which avoids some of the common drawbacks of region merging techniques. The algorithm has both statistical and computational advantages over known approaches. It authorizes the processing of 512×512 images in less than a second on conventional PC computers. Experiments are reported on thirty-five images of various origins, illustrating the quality of the segmentations obtained.