Open AccessMulti‐step radiographic image enhancement conforming to weld defect segmentation
Author(s) -
Dang Changying,
Gao Jianmin,
Wang Zhao,
Chen Fumin,
Xiao Yulin
Publication year - 2015
Publication title -
iet image processing
Language(s) - English
Resource type - Journals
SCImago Journal Rank - 0.401
H-Index - 45
eISSN - 1751-9667
pISSN - 1751-9659
DOI - 10.1049/iet-ipr.2014.0716
Subject(s) - artificial intelligence , computer vision , image segmentation , radiographic testing , segmentation , histogram , adaptive histogram equalization , mathematics , smoothing , pattern recognition (psychology) , image histogram , weighting , scale space segmentation , contrast (vision) , computer science , histogram equalization , image texture , image (mathematics) , welding , materials science , medicine , metallurgy , radiology
To improve the accuracy of automatic defect segmentation in radiographic non‐destructive testing and evaluation, the authors proposed a multi‐step radiographic image enhancement algorithm (MSRE) conforming to weld defect segmentation. In this algorithm, the requirement of defect segmentation is fully considered when enhancing a radiographic testing image. The first‐step enhancement is performed by linear weighting between an original radiographic image and its contrast‐limited adaptive histogram equalisation image. Anisotropic diffusion filtering is used for simultaneously smoothing the weighted image and preserving the defect edges very well. Then, the filtered image is enhanced by a fuzzy enhancement algorithm as the final‐step enhancement and hence, obtaining a new image with high contrast, high definition, and strong edge intensity. The authors compared MSRE with adaptive histogram equalisation, fuzzy enhancement, global contrast enhancement, and local contrast enhancement algorithms and evaluated its performance by using indicators such as image contrast, definition, edge intensity, and information entropy. Furthermore, the authors compared the segmentation results of the enhanced images to further study the algorithm's effect on weld defect segmentation. Experimental results reveal that the quality of enhanced images is significantly improved by MSRE, and the image enhanced by MSRE has an high relative segmentation accuracy (RSA) of more than 95%.