Show simple item record

dc.contributor.authorEhrich, Roger W.en_US
dc.contributor.authorFoith, J. P.en_US
dc.date.accessioned2013-05-28T20:43:12Zen_US
dc.date.accessioned2013-06-19T14:37:01Z
dc.date.available2013-05-28T20:43:12Zen_US
dc.date.available2013-06-19T14:37:01Z
dc.date.issued1977
dc.identifierhttp://eprints.cs.vt.edu/archive/00000819/en_US
dc.identifier.urihttp://hdl.handle.net/10919/20296
dc.descriptionOne lesson that has been learned from previous approaches to scene analysis is that local methods are insufficient for extracting reliable information about the contents of a scene. Two different procedures that have been tried in order to remedy this deficiency are the use of knowledge via a priori information and internal models and multilevel analysis based on hierarchies of representations such as cone systems. It does not seem appropriate to drive the very first levels of analysis by a priori knowledge. It is doubtful that it will be possible to use knowledge in a way general and versatile enough to direct low level processing, and there is a need for some powerful data driven mechanisms that might at a later stage invoke internal models. It would seem more appropriate to obtain some crude global information through glancing or planning at low resolution levels that can drive a more scrutinous analysis at high resolution levels. While hierarchal systems are therefore good, the way they are currently being constructed is not necessarily good. In this context the issue of low level representation becomes more and more important, and not enough attention has been paid to this issue. Even Marr's provocative ideas about his primal sketch do not go to a sufficient level of analysis, and it is felt that more of the workload should be thrown onto the first processing levels. In this paper is posited a comprehensive hierarchal data structure that requires no decisions and therefore no parameters for its construction. The technique does not require preselected windows, but rather uses context-dependent criteria. The data structure is versatile, easily computed, and invertible in the sense that the original image is completely recoverable.en_US
dc.format.mimetypeapplication/pdfen_US
dc.publisherDepartment of Computer Science, Virginia Polytechnic Institute & State Universityen_US
dc.relation.ispartofHistorical Collection(Till Dec 2001)en_US
dc.titleStructural Processing of Visual Informationen_US
dc.typeTechnical reporten_US
dc.identifier.trnumberCS77004-Ren_US
dc.type.dcmitypeTexten_US
dc.identifier.sourceurlhttp://eprints.cs.vt.edu/archive/00000819/01/CS77004-R.pdf


Files in this item

Thumbnail

This item appears in the following Collection(s)

Show simple item record