Skip to Main content Skip to Navigation

Appariement d'images par invariants locaux de niveaux de gris. Application à l'indexation d'une base d'objets

Cordelia Schmid 1
1 MOVI - Modeling, localization, recognition and interpretation in computer vision
GRAVIR - IMAG - Laboratoire d'informatique GRAphique, VIsion et Robotique de Grenoble, Inria Grenoble - Rhône-Alpes, CNRS - Centre National de la Recherche Scientifique : FR71
Abstract : This thesis concerns matching, a fundamental subject in computer vision. Matching covers a variety of problems such as matching two images or matching an image with a CAD model. Our approach allows objects to be matched if they are observed in complex scenes, partially occluded or seen from different viewpoints. The method is extended to image database consultation and object recognition. Our approach is based on a local characterization of the greyvalue signal. This characterization is calculated at particular «points of interest». These are detected automatically and are representative of the observed object. Therefore, the characterization obtained has a high information content. In addition, it is invariant to the similarity group of transformations in the image and allows images that have undergone such transformations to be matched. To first order, the similarity group absorbs variations of perspective viewpoint changes, so our representation is quasi-invariant and therefore robust to such transformations. The method has been applied to the retrieval of images from a large database. When there are many images there are typically many possible matches for any given point, so a robust statistical technique has been developed to find the corresponding image. To reduce the amount of computation required for a large database and make rapid retrieval possible, an indexing mechanism has been developed. Our image retrieval scheme has been applied to 3D object recognition from a single image. Each object is modeled by a set of images taken from different viewpoints chosen to be representative of the object. To obtain 3D information, the different aspects of the objects stored in the database are annotated with symbolic data. The trilinearity constraint allows this data to be localized in the image.
Document type :
Complete list of metadatas

Cited literature [24 references]  Display  Hide  Download
Contributor : Thèses Imag <>
Submitted on : Monday, February 23, 2004 - 6:11:11 PM
Last modification on : Monday, December 28, 2020 - 3:44:02 PM
Long-term archiving on: : Wednesday, September 12, 2012 - 3:50:32 PM


  • HAL Id : tel-00005019, version 1




Cordelia Schmid. Appariement d'images par invariants locaux de niveaux de gris. Application à l'indexation d'une base d'objets. Interface homme-machine [cs.HC]. Institut National Polytechnique de Grenoble - INPG, 1996. Français. ⟨tel-00005019⟩