Analysis of Image Segmentation for Radiographic Images

Digital image processing is used to identify regions in an image by using various segmentation methods. This paper gives a brief account on five of the different segmentation techniques namely region growing, watershed, thresholding, split and merge, k-means clustering methods highlighting the advantages as well as disadvantages of each of these methods. A modification of traditional region growing segmentation method is presented which automatically selects the seed points and grows the regions until all the regions in the image are segmented. The results of segmentation methods presented in the paper are not dependent on the kind of image to be segmented and these methods are used in segmenting industrial radiographic weld images in which several defects like porosity, lack of fusion, slag line, incomplete penetrations, and wormholes occur. The methods are evaluated on various types of images and efficiency of these methods in the detection of several weld defects is presented along with the experimental results. The evaluation of performance of these different segmentation methods on sample images is done on the basis of subjective criteria and conclusions are achieved. These methods are used to detect the flaws in an object by identifying the flawed region in the image. Due to this ability of region detection, it finds various applications in medical imaging, optical character recognition, computer vision, remote sensing, mobile robots and industrial radiography.