Performance comparisons of contour extraction algorithms

Abstract

Computer analysis of video images is a popular technique for noncontact measurement and inspection. These applications frequently require the identification and extraction of the contours of areas of interest in the image from which a variety of measurements can be made including location, orientation, and shape. Recent examples include detecting flaws in hot steel slabs, measuring surface defects in wood, finding growth traces in microbiology samples, and a host of applications in the area of robotics and flexible manufacturing. Several techniques for contour extraction are surveyed; a new algorithm is described and compared with previous methods in terms of number of calculations, memory requirements, and data structures. We show that the new algorithm performs significantly faster and is more robust at the expense of a small increase in memory. Due to its sequential nature, it is suitable for on-line execution during the raster scan of a TV sensor.