Recherche d'objets par appariement de graphes combinant contours et régions
Résumé
In the last decade, the 3D modeling of cities has become one of the challenges of multimedia search and an important focus in object recognition. We are interested in locating various primitives, especially windows, in Paris façades. An image of edges is readable by the human visual system, which performs a perceptual grouping and is thus able to catch the entities of the image.We tried in this paper to model this mecanism. Images are represented as adjacency graphs of contour segments, attributed with orientation and proximity information. To achieve the graph matching, we propose several variants of a new similarity based on sets of paths carried by the graphs, able to perform the perceptual grouping and robust to scale changes. The similarity between paths takes into account the similarity between sets of edges and the similarity between regions defined by these paths. The retrieval of images containing a specific object is performed thanks to a SVM or knn classifier. For object localization, we use a vote system from the paths selected by the matching algorithm. Our results show that we are able to extract the windows from images of façades from examples.
Origine : Fichiers produits par l'(les) auteur(s)
Loading...