Color image segmentation using region adjacency graphs

An important aim of image analysis is to segment an image into parts that correlate strongly with real work objects or surfaces. In this paper, we present an algorithm for color image segmentation that uses a transformation based on region adjacency graphs. We examine partitions of region adjacency graphs that make it possible to pass rapidly from an initial representation of an image that uses a large number of regions to a final segmentation that is composed of appropriately few regions. After an initial segmentation process we create an associate adjacency graph, and partition this graph in two ways: by growth on graph vertices using region growing, and by a watershed transformation of a contour graph.