DESCRIPTIVE LOGIC OF SEARCHING OBJECTS ON IMAGES

Aleksandr V. Kuchuganov (Aleks_KAV@udm.ru)

Kalashnikov Izhevsk State Technical University

The most popular approaches to the problem of searching objects on images are: a linguistic approach, within which syntactic recognition of objects of strictly predetermined structure is carried out; artificial neural networks; descriptive image algebras; approaches that describe objects using predicate logic; CBIR (Content-Based Image Retrieval) technologies, based on descriptive logics (DL). The article presents an ontological approach based on descriptive logic with an extension to the spatial domain of data represented in the form of attributive graphs. The process of image analysis is controlled by a strategy containing: a preliminary morphological analysis stage; stage of the formation of a hypothesis about the category of the found object; stage of confirmation of the hypothesis by a logical conclusion about the class of the object. In the course of the analysis, a decision tree is constructed on the categories of objects. After selecting a hypothesis, a derivation tree is formed about the correspondence of the object to any DL-definition from this category. With an unsatisfactory degree of similarity and differences, a transition to another branch of the decision tree about categories occurs. Examples of the work of the image analysis system are shown.

attributive graph, cluster analysis, descriptive logic, granular ontology, thesaurus, decision tree, output tree, degree of similarity and differences, graph of image objects.

Back