Accession Number : ADA192036

Title :   Linguistic Definition of Generic Models in Computer Vision,

Corporate Author : ROYAL SIGNALS AND RADAR ESTABLISHMENT MALVERN (ENGLAND)

Personal Author(s) : Fretwell, P ; Goillau, P J

PDF Url : ADA192036

Report Date : Dec 1987

Pagination or Media Count : 12

Abstract : The method has been developed that can take a human description of an object's spatial appearance and produce a PROLOG representation. The object's appearance is currently in terms of an edge map and the English descriptions are stylised accounts of the salient features and combinations of features found in this representation. At present the translation is performed by hand. However, suggestions are made on how this process can be automated. A prototype translator has been implemented. The PROLOG model is expressed as a hierarchy about the object's appearance, terminating in plausible low-level image primitives. A way is proposed of matching the hierarchy against an image for object recognition in isolation from its background. This reduces the search space of features and feature combinations that the matcher has to consider, so avoiding some of the combinations that the matcher has to consider, so avoiding some of the combinatorial problems when using PROLOG. Extensions using fuzzy logic to deal with uncertain image data and the vagueness of natural language are discussed.

Descriptors :   *NATURAL LANGUAGE, *IMAGE DISSECTION, *ARTIFICIAL INTELLIGENCE, COMBINATORIAL ANALYSIS, COMPUTERS, ENGLISH LANGUAGE, HIERARCHIES, HUMANS, IMAGES, ISOLATION, LOGIC, LOW LEVEL, PROTOTYPES, RECOGNITION, TRANSLATORS, VISION, IMAGE PROCESSING, COMPUTER APPLICATIONS, GREAT BRITAIN

Subject Categories : Cybernetics

Distribution Statement : APPROVED FOR PUBLIC RELEASE