Publication

Abstract : Measurement of surface profiles of manufactured components are of critical importance and are traditionally carried out using contact probes and using a few optical methods in the past few decades. With advances in digital image processing techniques and computational power, it is advantageous to use digital optical metrology for non-contact measurements—a field still under active development. For instance, for two-dimensional form measurements, essential steps in roundness characterization include capturing a good digital image of top-view of a sample and processing the image to compute boundary-edge. Accuracy and difficulty of this computation is influenced by textural background noise—a problem of significant complexity in image segmentation and measurement, and methods specific to applications are being developed. For experimental setups with backlighting, the background surrounding the region of interest is noise-free. However, for images with reflected lighting, noise due to background surface and dust must be dealt with. Working with an image of a top-view of a cylinder, we define a new measure of background noise using Canny-edge output for this purpose and present a method of characterizing the background surface noise. We study the effect of variation in the Canny-edge parameters—Gaussian kernel width (sigma) and threshold—on this background noise and arrive at a transition threshold for which the background noise is zero for various sigma values. For the chosen experimental conditions, we find that the background noise is absent for thresholds above 0.4852 for any sigma, and there is a critical sigma of 2.5 at which the transition threshold is minimum at 0.0777. These values give valuable insights on the choice of these parameters and on the robustness of the estimated boundary-edge.