Zurück zur Übersicht

Query by Sketching (QbS)

Titel Englisch Query by Sketching (QbS)
Gesuchsteller/in Schuldt Heiko
Nummer 117800
Förderungsinstrument Projektförderung (Abt. I-III)
Forschungseinrichtung Fachbereich Informatik Departement Mathematik und Informatik Universität Basel
Hochschule Universität Basel - BS
Hauptdisziplin Informatik
Beginn/Ende 01.10.2007 - 30.09.2009
Bewilligter Betrag 195'174.00
Alle Daten anzeigen

Keywords (9)

digital libraries; content-based image retrieval; interactive paper; semantic annotation; relevance feedback; Content-based Image Retrieval (CBIR); Image Similarity Search; Information Retrieval; Human-Computer Interaction

Lay Summary (Englisch)

Lay summary
Increasing numbers of large image collections are available in Digital Libraries. These collections significantly impact the way people access and use information, both in their business and private lives. The Web 2.0, for instance, has generated many so-called social networks such as flickr and Wikimedia Commons to allow people to share images. Users can manually tag uploaded images with textual meta data describing the content, known as collaborative tagging. However, it is almost impossible to associate an image with a comprehensive set of objective tags based on user perception in order to support all kinds of queries. This leads to rather poor retrieval quality when issuing a keyword search. Other application domains are facing similar problems. In healthcare, for instance, more and more modalities such as CT (Computed Tomography) and MRT (Magnetic Resonance Tomography) produce high resolution images which are stored in a patient’s electronic health record without any information on their actual content.

The information retrieval community has addressed the problem of advanced information access for image data beyond textual keyword queries by making use of the image content itself (Content-based Image Retrieval, CBIR). Despite strong technical support, CBIR systems are still not in widespread use. The reason for this is twofold. First, content-based retrieval (”similarity search”) requires a query image to start with that is sufficiently close to the final result, i.e. that precisely expresses the user’s information need. Without such a query image, it is difficult or nearly impossible to have good retrieval quality, even if powerful relevance feedback mechanisms are available.
Second, in many applications users are particularly interested in certain regions of the picture. Assume, for instance, a CT or MRT image of a patient with a tumor. From the physician’s point of view, similarity should be restricted to the region of the image where the tumor is detected, not to the overall image. However, current approaches either consider similarity in global terms or apply conventional segmentation techniques which do not take into account any application semantics.

The goal of the QbS project is to combine CBIR and novel interactive paper interfaces to address these problems by providing a naturalistic interface that will significantly lower the barriers for content-based image retrieval in a variety of applications. Users will be able to sketch their query on paper through a combination of drawing similar shapes, recognised symbols, annotations and command gestures. The interactive paper interface to CBIR will also allow the user to dynamically highlight regions of interest. Since the user-defined region selection is much more flexible than automatic segmentation-based region detection, it is assumed to better match the application-specific needs of CBIR users.
Direktlink auf Lay Summary Letzte Aktualisierung: 21.02.2013

Verantw. Gesuchsteller/in und weitere Gesuchstellende


Verbundene Projekte

Nummer Titel Start Förderungsinstrument
126829 PAD-IR: Paper-Digital System for Information Capture and Retrieval 01.10.2009 Projektförderung (Abt. I-III)