Питання синтезу дискретних зображень в задачах розпізнавання образів

Abstract

For solving a class of pattern recognition (classification) of images generally face the following situation – currently accumulated a significant amount of algorithmic and methodological tools, which solve some particular tasks, subtasks (description or performance) images characteristics (structural elements) in images and more, however, there is no uniform methodology for their effective use, and there is no simple, universal methodology information (indicative) description of the image. Today, there are a number of approaches, methods and algorithms for the selection of features in images and software packages for their implementation. However, there remains the problem of finding a system of optimal (in a sense, for the current task) features, that is, the search for such properties of images (definition and fixation of the feature space) in the space of which classification (recognition) would be possible and not very difficult (cost-effective) task. The use of existing algorithms and methods for this problem becomes possible only in the presence of methods that would be based on the results of different systems, allowed to allocate the system of features that are the most qualitative within the current problem. Moreover, for each practical problem of image classification, the feature systems that are relevant at this stage (important relative to a fixed problem or class of problems) are usually different and need to be redefined. Hence, it becomes obvious the relevance of this study, in relation to the important task of finding the optimal (in a certain sense) feature systems. Often the problem of finding optimal feature systems is reduced to the problem of minimizing the original image description. However, this applies only to the case when the optimal system of features is among the sets of features that define the description of images, which is usually only an assumption. This work offers a way to minimize the initial description of discrete images, which allows us to build a minimal description of the image of an arbitrary structure on the basis of the concept of a T-reference set. The paper also introduces the concept of a T-reference set, and based on it is proposed to use the data sets as features of discrete images.