JPL Technical Report Server

Trainable Cataloging for Digital Image Libraries with Applications to Volcano Detection

Show simple item record

dc.contributor.author Burl, M. C. en_US
dc.contributor.author Fayyad, U. M. en_US
dc.contributor.author Perona, P. en_US
dc.contributor.author Smyth, P. en_US
dc.date.accessioned 2004-10-03T05:04:59Z
dc.date.available 2004-10-03T05:04:59Z
dc.date.issued 1995
dc.identifier.citation IEEE Transactions on Pattern Analysis and Machine Intelligence en_US
dc.identifier.clearanceno 95-1107
dc.identifier.uri http://hdl.handle.net/2014/31428
dc.description.abstract Users of digital image libraries are often not interested in image data per se but in derived products such as catalogs of objects of interest. Converting an image database into a usable catalog is typically carried out manually at present. For many larger image databases the purely manual approach is completely impractical. In this paper we describe the development of a trainable cataloging system: the user indicates the location of the objects of interest for a number of training images and the system learns to detect and catalog these objects in the rest of the database. In particular we describe the application of this system to the cataloging of small volcanoes in radar images of Venus. The volcano problem is of interest because of the scale (30,000 images, order of 1 million detectable volcanoes), technical difficulty (the variability of the volcanoes in appearance) and the scientific importance of the problem. The problem of uncertain or subjective ground truth is of fundamental importance in cataloging problems of this nature and is discussed in some detail. Experimental results are presented which quantify and compare the detection performance of the system relative to human detection performance. The paper concludes by discussing the limitations of the proposed system and the lessons learned of general relevance to the development of digital image libraries. en_US
dc.format.extent 2416219 bytes
dc.format.mimetype application/pdf
dc.language.iso en_US
dc.subject.other digital image libraries, trainable cataloging en_US
dc.title Trainable Cataloging for Digital Image Libraries with Applications to Volcano Detection en_US


Files in this item

This item appears in the following Collection(s)

Show simple item record

Search


Browse

My Account