Evolutionary Region Growing for Image Segmentation

摘要

Image segmentation is important stage in image processing. Segmentation by region growing is a fast, simple and easy to implemented, but it suffers from three disadvantages: the threshold problem, initialization of germs, and the processing order of the neighboring pixels of germs. Evolutionary algorithms are particular methods for optimizing functions; they have a great ability to find the global optimum of a problem. In this paper, we used evolutionary algorithms to get over the three problems of region growing. We have proposed a segmentation method based on region growing and evolutionary algorithms. The proposed approach is validated on 1000 synthetic images (500 images with 2 regions and 500 images with 3 regions). Thus, our method is validated on five unsupervised evaluation criteria: Inter-region contrast Vinet criterion and compared with two methods: split and merge method and Kmeans algorithm. The results show the good performance of this approach.