Segmentation interactive d'images fixes et de séquences vidéo basée sur des hiérarchies de partitions

Abstract : In multimedia applications, the diversity of treated images makes automatic segmentation very difficult. Our approach aims at obtaining good quality segmentations by allowing a minimum amount of user interaction. In order to achieve high flexibilty and fast response times, a video sequence is represented as a set of nested partitions. All possible contours in the image are detected, together with an estimate of their importance. The interactive step allows the user to choose the contours that correspond to the objects of interest. In this way,multiple segmentations are possible without significant recalculations. Initially, various hierarchies associated to morphological floodings are studied, together with different mechanisms allowing the introduction of a priori knowledge when this knowledge is available. Later on, the notions presented for still images are generalized to the case of video sequences. A single hierarchy representing a video sequence shot is calculated. Several interaction tools are also proposed, that allow the user to manipulate the hierarchie in an intuitive manner. The tree structure chosen to represent the hierarchie allows for an efficient manipulation of the information, and the results of the interaction are perceived by the user as being immediate.
Document type :
Theses
Liste complète des métadonnées

https://pastel.archives-ouvertes.fr/pastel-00003264
Contributor : Ecole Mines Paristech <>
Submitted on : Wednesday, January 9, 2008 - 8:00:00 AM
Last modification on : Tuesday, September 29, 2015 - 10:32:46 AM
Document(s) archivé(s) le : Wednesday, September 8, 2010 - 5:56:01 PM

Identifiers

  • HAL Id : pastel-00003264, version 1

Collections

Citation

Maria Fransisca Zanoguera Tous. Segmentation interactive d'images fixes et de séquences vidéo basée sur des hiérarchies de partitions. Mathematics [math]. École Nationale Supérieure des Mines de Paris, 2001. English. ⟨pastel-00003264⟩

Share

Metrics

Record views

338

Files downloads

877