The performance of some implicit region-based active contours in segmenting and restoring welding radiographic images

Abstract

Several domains are based on image processing and analysis. One of them is the radiographic inspection which is used in Non Destructive Testing (NDT). Active contours, snakes or deformable models are powerful techniques in image segmentation and restoration. According to the term related to the input data (image to be treated) those functionals are ranked on two categories: edge-based models and region-based models. Previous studies point out the advantages of the regionbased models over edge-based models. In this paper, we discuss and we summarize the strengths and weaknesses of four implicit region-based active contour models named: Piecewise Constant PC, Piecewise Smooth PS, Local Binary Fitting LBF and Global Local fitting energy GLF. After performing several experiments, we have concluded that all the models perform well with homogeneous images. On the contrary when images are strongly inhomogeneous, the models based on global (PC) or local (LBF) statistic intensity fail to segment such images. The PS model with its great advantage in preserving the contours has, as a drawback, the high CPU time consuming. The combination of local and global statistic image intensity gives to the GLF model the ability to better deal with such images in less CPU time.