Résumé
This paper deals with "on the fly" segmentation on sliding images. We describe a quality control industrial application for which that type of image analysis is an elegant solution. Local and global merging techniques on the region adjacency graph are discussed. Only local merging allows us to process an uncompletely known graph. Global merging on the entire graph and "on the fly" local merging give similar segmentation results, and yet the iatter is iaster. Keywords: image analysis, region adjacency graph, "on the fly" processing, quality control. the choosen vertex and its neighbourhood is sufficient in order to start the merging process, global merging: we take into account all the pairs of adjacent vertices in the whole graph; if some of these pairs can be merged then we select the best if any [PCCBSl]. The entire graph is required when the merging process starts. In the following, we present these two tech-niqnes, and we detai! some advantages of ln-cal merging. We also describe an industrial application in which local merging was used. It allows us to do "on the fly" processing.