Cracklike defects detection and sizing from co-occurrence matrices

The inspection of austenitic welds used in nuclear field with ultrasounds poses problems in interpretation: strong grain noise makes difficult the detection of the crack top and the crack bottom. Since corresponding echoes enable the defect sizing, defect sizing also becomes difficult. The formation of 2D images (BSCAN), and their processing enable an increase in the effectiveness of testing. We present a segmentation method, based on co-occurrence matrix, which separates defects zones and noise zones. Examples of segmentation improvement applied to artificial defects are presented