Dynamic adaptive data structures for semantic analysis and synthesis of video information

Abstract

The wide range of image processing methods such as wavelets and fractal decomposition or texture analysis usually miss the vital components — the semantic structure.In our approach to the structuring of video-data, which was realised in SAI (Semantic Analysis of Images) software package, we apply the adaptive dynamic data structure for object fitting hierarchical analysis of video-data. This package presents the new software tools to reveal the interrelated network of context independent semantics from the initial data structure.The scientific basis of our approach is the localisation of semantically important areas through the new rating principle and following iterative synthesis of hierarchical trees to create the coherent structure of selected fragments. The investigations lead us to the idea that the distribution of the fragments (F) between the levels (ι) of this tree correlates with the growth law F ι≈ ι−0.618, an analogy to the empirical Zipf laws.Adaptive dynamic data structure, as the result of this process, contains the image fragments which are essentially important for the following processing.Most evident application of semantic decomposition is the preliminary structurization of video-data for subsequent application of object identification routines and target-oriented discriminating compression of images.KeywordsCoherent StructureSource ImageImage RepresentationOptic Character RecognitionRating PrincipleThese keywords were added by machine and not by the authors. This process is experimental and the keywords may be updated as the learning algorithm improves.