Comparison of Neural Network and Markov Random Field Image Segmentation Techniques*
The interpretation of data from nondestructive evaluation (NDE) techniques is a tedious and time-consuming manual process that is subject to such random variables as scan quality, and inspector expertise and fatigue. The authors are researching methods to automatically recognize defects in ultrasonic images of aircraft structures. A typical wing skin image with an annotated defect is shown in Figure 1. Our ultimate goal is to reduce total fabrication time and improve inspection reliability.