Descriptive Image Analysis: Part III. Multilevel Model for Algorithms and Initial Data Combining in Pattern Recognition

• Published in: Pattern recognition and image analysis Vol. 30; no. 3; pp. 328 - 341
• Main Authors: Gurevich, I. B., Yashina, V. V.
• Format: Journal Article
• Language: English
• Published: Moscow Pleiades Publishing 01.07.2020; Springer Nature B.V
• Subjects: Algorithms; Analysis; Automation; Computer Science; Data analysis; Decision analysis; Decision making; Fragments; Image analysis; Image Processing and Computer Vision; Mathematical analysis; Mathematical Theory of Images and Signals Representing; Object recognition; Optimization; Pattern Recognition; Processing; Recognition; Representations; Spatial data; Standardization; Two dimensional analysis; Understanding; descriptive image models; descriptive image representations; dual representation of images; combinatorial structures of local neighborhoods; image analysis; image models; image recognition algorithms; image representations; mathematical theory of image analysis; models of image analysis procedures; support sets; multiple classifiers; descriptive image analysis
• ISSN: 1054-6618; 1555-6212
• DOI: 10.1134/S1054661820030086

This is the third article in a series of publications devoted to the current state and future prospects of Descriptive Image Analysis (DIA), one of the leading and intensively developing branches of modern mathematical theory of image analysis. The fundamental problem of computer science touched by the article is the automation of extracting from images of information necessary for intellectual decision-making. A new class of models for the image analysis and recognition process and its constituent procedures is introduced and described – a multilevel model of image analysis and recognition procedures (MMCAI) – which is based on the joint use of methods of combining algorithms and methods of combining fragmentary initial data – partial descriptions of the object of analysis and recognition – an image. The architecture, functionality, limitations, and characteristics of the MMCAI are justified and defined. The main properties of the MMCAI class are as follows: (a) combining the fragments of the initial data and their representations and combining algorithms at all levels of image analysis and recognition processes; (b) the use of multialgorithmic schemes in the image analysis and recognition process; and (c) the use of dual representations of images as input data for the analysis and recognition algorithms. The problems arising in the development of the MMCAI are closely related to the development of the following areas of the modern mathematical theory of image analysis: (a) algebraization of image analysis; (b) image recognition algorithms accepting spatial information as input data; (c) multiple classifiers (MACs). A new class of models for image analysis is introduced in order to provide the following possibilities: (a) standardization, modeling, and optimization of Descriptive Algorithmic Schemes (DAS) that form the brainware of the MMCAI and processing heterogeneous ill-structured information – dual representations – spatial, symbolic, and numerical representations of the initial data; (b) comparative analysis, standardization, modeling, and optimization of different algorithms for the analysis and recognition of spatial information. The fundamental importance of the results of these studies for the development of the mathematical theory of image analysis and their scientific novelty are associated with the statement of problems and the development of methods for modeling the processes of automation of image analysis when ill structured representations of images, including spatial data proper – images and their fragments, image models, incompletely formalized representations, and subsets of combinations of these representations – are used as the initial data. The introduction of the MMCAI as a standard structure for representing algorithms for the analysis and recognition of two-dimensional information and dual representations of images allow one to generalize and substantiate well-known heuristic recognition algorithms and assess their mathematical properties and applied utility.