Oversegmentation control for inexact graph matching: first results
Isabelle Bloch | Consularo, Luís Augusto | Cesar Jr, Roberto M. | de Figueiredo, Luiz Henrique
An interactive image segmentation method based on structural pattern recognition has been recently introduced. A model graph is generated from an oversegmentation of the image and from traces provided by the user. An input graph is generated from the oversegmented image. Image segmentation is then obtained by matching the input graph to the model graph. An important problem that should be addressed is how to control the size of the input graph. This size is given by the number of regions provided by the oversegmentation. To address this problem, we propose to control the maximum number of regions provided by the oversegmentation by using watershed with markers. The markers are given automatically by using two approaches: quadtrees and centroidal Voronoi diagrams. Results on real images are discussed.