Segmentation of Betatron SEA 7 Radiographic Images with Computed Radiography on Steel Specimen

Abstract

Radiographic testing is carried out to find out defects or discontinuities in a material that generally uses x-rays. SEA 7 is one of the new generations of betatron that produces high-energy x-ray radiation. In this study, an image from betatron with an energy of 7 MeV was obtained. The image is denoised using a discrete wavelet transform and segmented with a local thresholding method to see defects in the radiographic image. The Haar and Daubechies are a family of wavelets whose performance is tested on this image. The results show that the Daubechies family at the first decomposition level is more effective in reducing noise, as shown by the PSNR (Peak Signal to Noise Ratio) value of 32.62 dB for sample 1, namely the image with incomplete penetration defects, and 37.15 dB for sample 2, namely the image with the type of lack of root penetration defect. In the segmentation test, Niblack's segmentation performance after the morphological operation process reached an MMo (Misclassified Area Mutual Overlap) value of 86.65% for sample 1 and 68.09% for sample 2. The second local thresholding segmentation performance, namely Sauvola after the morphological operation process, reached an MMo value of 84.97% for sample 1 and 52.44 % for sample 2. The window used for segmentation exceeds the cropping size, so these two methods are not suitable to be applied on betatron radiographic images.