Deblurring Sinograms Using a Covolutional Neural Network to Achieve Fast X-ray Computed Tomography Scanning

• Published in: E-journal of Nondestructive Testing Vol. 25; no. 2
• Main Authors: Yuki, Ryo, Ohtake, Yutaka, Suzuki, Hiromasa
• Format: Journal Article
• Language: English
• Published: 01-02-2020
• ISSN: 1435-4934; 1435-4934
• DOI: 10.58286/25355

X-ray computed tomography (CT) allows for visualization of the interior of solid objects in a non-destructive and non-invasive
manner. However, producing high-precision measurements takes a long time because thousands of sharp transmission images
are required to reconstruct CT volumes. To address this problem, we propose a CT measurement method based on Convolutional
Neural Networks (CNNs) that yields sharp transmission images by deblurring blurry ones. In this method, first, blurry images are
obtained in a short measurement time, then they are deblurred by CNNs with fine-tuning and integrated by linear interpolation.
This method shortens the measurement time indirectly because the process related to the CNNs is fast with GPUs and the blurry
images with low levels of noise intensity do not require a long time. Besides, the fine-tuning may improve the output images’
sharpness. According to our experimental results, the proposed method is fast and can maintain the quality of data to a certain
extent.