Defects extraction of X-ray weld image based on SUSAN

CHEN Qiang1， XUN Yi1， CUI Di2， BAO Guan-jun1， YANG Qing-hua1
（1. Key Laboratory of E&M， Ministry of Education & Zhejiang Province， Zhejiang University of Technology，Hangzhou 310032， China；
2. School of Biosystems Engineering and Food Science， Zhejiang University， Hangzhou 310027， China）

Abstract： Aiming that X-ray image of weld contains low contrast，uneven gray distribution，much image noise and more dynamic fuzzy and other shortcomings which could lead to difficult to extract the weld defects，an algorithm of weld defects extraction was put forward based on small univalue segment assimilating nucleus（SUSAN）. Firstly，the region of interested（ROI） area was extracted to reduce the amount of processed data and median filter of this area was done，then the entry point and exit point of weld defects of each line were found by SUSAN algorithm. By combing mathematical morphological operation，the noise of isolated spot and discontinuous points was filtered to extract the accurate position of defect area. Through the VC++ 6.0 platform for experiments to test some common defects，the correct segmentation rate is about 87% compared with artificial observation. The results show that the contrast is too low and the noise is too much of weld defects are the main reasons that could not segment the weld defects correctly.
Key words： X-ray weld image region of interested（ROI）； weld defect extraction； small univalue segment assimilating nucleus（SUSAN）； mathematical morphological