Image Representations for Ranking and Classification

Josip Krapac 1, 2
1 Equipe Image - Laboratoire GREYC - UMR6072
GREYC - Groupe de Recherche en Informatique, Image, Automatique et Instrumentation de Caen
2 LEAR - Learning and recognition in vision
Inria Grenoble - Rhône-Alpes, LJK - Laboratoire Jean Kuntzmann, INPG - Institut National Polytechnique de Grenoble
Abstract : This thesis concerns the tasks of image re-ranking and image classification. These tasks are solved by learning statistical models given a representation of visual content of the image and a similarity measure between images. Here we aim to improve performance of the tasks by extending the bag-of-words image representation, while using existing statistical models and similarity measures between images. We adapt the image representation according to a given task. First we explore the task of image re-ranking, whose goal is to re-order the images retrieved by a text query such that images relevant to a query are ranked above non-relavant ones. Inspired by text re-ranking methods we developed a query-relative image representation that depends on the visual content of the image, but also on the query used to retrieve it. Next, we adapt the representation for the task of image classification, which aims to assign one or more labels to an image that is related to the content of the image. We have adapted the representation by learning a visual vocabulary specifically for the classification task. We have also introduced a new representation that encodes the information about spatial layout of image parts in much more compact manner than currently used representations that encode the spatial layout. All developed image representations are compact, fast to construct and already perform very good with linear models. We show marked improvements on several stan- dard and challenging datasets with respect to state-of-art-methods. For image classification and image re-ranking tasks we have shown that adapting the representation to the task improves the performance.
Complete list of metadatas

Cited literature [107 references]  Display  Hide  Download

https://tel.archives-ouvertes.fr/tel-00650998
Contributor : Josip Krapac <>
Submitted on : Monday, December 12, 2011 - 3:50:53 PM
Last modification on : Tuesday, February 5, 2019 - 12:12:44 PM
Long-term archiving on : Sunday, December 4, 2016 - 9:20:27 PM

Identifiers

  • HAL Id : tel-00650998, version 1

Citation

Josip Krapac. Image Representations for Ranking and Classification. Computer Vision and Pattern Recognition [cs.CV]. Université de Caen, 2011. English. ⟨tel-00650998⟩

Share

Metrics

Record views

1136

Files downloads

1880