Comparison of segmentation techniques for detection of defects in non-destructive testing images

Steam generator tubes and Obscured pipe lines like sewers, water mains have to be checked for their current condition. Cracks and defects are a strong indicator for the condition of a pipe. Electromagnetic nondestructive tests are important and widely used within the field of nondestructive evaluation (NDE). Magnetic Flux Leakage (MFL) has grown into a crucial method for inspection of pipelines and tubing in order to prevent long-term failures. Digital image processing techniques open the opportunity to accelerate the image analysis process, which may ease the operator from a lot of tedious task. An affordable way to detect those cracks is to take images of the pipeline and use image processing techniques to detect defects in these images. The Magnetic flux leakage images obtained from simulation software COMSOL multiphysics 4.3a are used for this work. Automatic segmentation is an important technique in the image processing. The basic idea of segmentation is to automatically select on gray-level values for separating object from the background. In this work, median filter is used to pre process the raw NDT image and three segmentation techniques are performed to segment the defect from the defective steam generator tube images. The performance evaluation of three segmentation algorithms namely region growing, minimum error thresholding and Morphological segmentation method for Non-Destructive testing (NDT) are performed and compared. Region growing technique is performed well for almost all MFL images.