Розроблено перетворення з узагальненими гребінчастими масштабними та вейвлет-функ-
ціями, що відрізняються від інших методів вейвлет-перетворення аналізуючими функціями. Ці функції характеризуються лінійчатим спектром, що призводить до низьких обчислюваль-
них витрат при визначенні на зображенні границь області структурної текстури, необхідних
для досягнення мети геометричних розмірів.
Запропоновані перетворення застосовано для локалізації судин на ангіограмах, що дозволило
скоротити час обробки цих зображень
Разработаны преобразования с обобщенными гребенчатыми масштабными и вейвлет-функциями, отличающиеся от используемых методов вейвлет-преобразования анализирующими функциями. Эти функции характеризуются линейчатым спектром, что приводит к низким вычислительным затратам при определении на изображении границ области структурной текстуры, требуемых для достижения цели геометрических размеров. Предложенные преобразования применены для локализации сосудов на ангиограммах, что позволило сократить время обработки этих изображений
When designing systems for computer recognition of visual images, segmentation of the processed class of images is an important stage. For many applied problems, the boundaries of objects, filled with structural texture often must be determined on an image. Existing methods for solving such problems are characterized by large computational cost. An alternative is using a parallel analysis of the spectral composition of the structural texture images for reducing image processing time and applying wavelet analysis for selecting objects of a given size. To analyze the image content in several frequency bands, it is advisable to use solutions of two-scale difference equation in the space of generalized functions. The solution of this equation for most sets of its coefficients can be obtained only approximately using the method of successive approximations. In the frequency domain, successive approximations of the solution of two-scale difference equation are characterized by a line spectrum and allow to perform parallel spectral analysis of structural texture images. Therefore, it is proposed to use these functions as analyzing functions of transforms with generalized comb scaling and wavelet functions to reduce image processing time due to replacing several processing levels by one. The proposed transforms were used in developing the information technology for vessel segmentation on angiograms. By applying the transform with the generalized comb wavelet function in the vessel localization, several processing levels were replaced by one. The latter is achieved by the fact that the convolution with the generalized comb wavelet function is similar to using a set of bandpass filters. As a result of experiments, it was shown that using the developed information technology provides vessel localization quality, required for making diagnostic decisions. In this case, the processing time is reduced by 43%, and the characteristics of detection reliability of vessel pixels are changed as follows: type I error probability is reduced by 1.22 times, and type II error probability is increased by 1.14 times.
