Description of image content by means of graph grammars

Abstract

This paper presents an idea for partial bottom-up parse of image content by use of an attributed graph grammar, in order to achieve effective high-level representation of knowledge contained in image. Terminal nodes of the proposed grammar are formed by image objects (points, lines, and objects detected by classifiers) and areas detected in image by various image processing and segmentation methods. Based on attributes of terminal nodes, each production rule creates derived attributes for high-level representation of lower-level knowledge. Graph that is parsed by graph grammar is constructed in process of knowledge extraction by application of segmentation and image processing algorithms. Created graph is then processed by sequential application of graph grammar rules. Left side of rules is detected by isomorphism detector, and consequent rewrite is performed by rule with highest priority. A part of rewrite process is represented by processing of evaluations of vertices and edges, that describe various properties of objects and their relationships. Further in the paper we present example of attributed graph grammar application in order to describe image content.