Page 10 - КЛАСИФІКАЦІЯ ЛОКАЛІЗОВАНИХ ДЕФЕКТІВ ЗВАРНИХ ШВІВ НА РЕНТГЕНІВСЬКИХ ЗОБРАЖЕННЯХ ТРУБ
P. 10
10
contrast for normalization of diagnostic images on the basis of the logarithmic type
model, complementary methods of processing weld images were developed, which in
combination with the segmentation of the weld zone and the segmentation of defects
made it possible to detect defects in accordance with the sensitivity of X-ray control.
An algorithm for restoring the input image with increased resolution, which does
not require the inversion of large matrices, has been developed. The algorithm is
based on the search for a signal with minimal energy and the use of the alternative
properties of the signal reconstruction error. The optimal reconstructed signal is
selected by the minimum length of the envelope curve of the signal, which reduces
the error of image reconstruction.
An algorithm for robust image filtering based on impulsive outlier localization
has been developed, which eliminates noise with a mixed distribution and can be
used to determine model parameters during image segmentation and calculation of
relevant features.
A number of developed methods for defect segmentation on digital radiographic
images of welded joints and calculation of their diagnostic features is described.
Characteristics of allowed defects are divided into absolute and relative. The linear
dimensions of the defects, their number in the weld and the distance between them
are absolute characteristics. The relative characteristics include the ratio of the linear
dimensions of the defects to the unit length or thickness of the weld.
A method of localization and segmentation of images based on a piecewise
linear model of the image intensity function has been developed in order to highlight
objects of interest. With the help of a multi-scale relevance function, one can locate
objects of different sizes. This method can be used to localize the groove standard on
radiographic images of welds, for segmentation of pores on the surface of coatings,
tracking the joint of two elements in the system of automatic control of their electron
beam welding, in medical radiography.
The approach to segmentation of radiographic images based on processing by
masks and morphological operators is proposed. The efficiency of its application for
segmentation of the weld zone, weights and non-welds is shown.