Sizing of prominent features in the images of important industrial materials using image segmentation

Abstract

A program has been written in Matlab for image segmentation based on user adjustable grayscale thresholding that can be used not only to locate and select important features in a given image but also count the number of pixels in the selection thereby sizing the feature(s) as a fraction of the total area occupied by the image. The program has been validated against ground truth realities of some locally fabricated samples in which our results indicate that if the grayscale threshold is kept fixed for all the fabricated samples in the program, it correctly selects and quantifies at least 85% of the actual required regions of interest in a given image or gives less than 4.6% false alarm rate. However, these numbers improve to about 97% and less than 0.35% respectively if the grayscale threshold sensitivity is allowed to be adjusted. Also, our program works well without the need of any smoothening or eroding morphological operations and is independent of image size. Therefore our program can be applied over standard reference images of nuclear fuel for their defect sizing due to chipping, cracks, voids, reshaping and other features such as material accumulation etc. The classes of fuel may range from fresh to post-irradiated, from light water to fast breeder reactors and from pellets to annular and vibro-compacted fuel.