computer vision, image processing, image contrast, frequency filtering, neural networks
Abstract
The article deals with the preprocessing of images by the computer vision system to find possible defects in transparent objects of complex shape made of amorphous materials. It is not always possible to obtain high-quality images with high contrast for such objects due to the small difference in the refractive indices of the product materials and the defect. The previously developed defect detection method is based on a modern neural network architecture and shows that image quality and contrast are critical indicators for effective defect detection. Therefore, the authors apply a special image frequency filtering technique to increase the contrast. The technique is based on dividing the image into narrow bands located perpendicular to the intensity gradient of the detail components image. One-dimensional forward Fourier transform, frequency filtering, and inverse Fourier transform are used for each band. Processing of a real image of the PS-70E (U70) insulator shows that the use of such frequency filtering reduces the contrast in the area of the components image and increases the contrast in the area of the defect image against a contrasting background. This property enables either identifying smaller size defects, or using images with resolutions up to and including 1024x1024 pixels, which can be useful when implementing computer vision systems in real industrial conditions.
Author Biographies
Anton Veniaminovich Korzhov, South Ural State University, Chelyabinsk
Dr. Sc. (Engineering), Associate Professor, Professor of the Department of Power Plants, Networks and Power Supply Systems, Vice-Rector for Research
Vladimir Anatol'evich Surin, South Ural State University, Chelyabinsk
Cand. Sc. (Engineering), Senior Lecturer, Department Mathematics and Programming
Petr Vladimirovich Lonzinger, South Ural State University, Chelyabinsk
Cand. Sc. (Engineering), Associate Professor, Department of Power Plants, Networks and Power Supply Systems
Valeriy Ivanovich Safonov, South Ural State University, Chelyabinsk
Cand. Sc. (Physics and Mathematics), Associate Professor, Department Power Plants, Networks and Power Supply Systems
Yaroslav Viktorovich Bushmelev, South Ural State University, Chelyabinsk
Engineer, Department of Scientific and Innovative Activities
Kirill Nikolaevich Belov, South Ural State University, Chelyabinsk
Postgraduate Student, Assistant of the Department Optoinformatics
