Computer vision algorithms versus traditional methods in food technology: The desired correlation

Abstract

Active Contours represent a common Pattern Recognition technique. Classical active contours are based on different methodologies (variational calculus, dynamic programming and greedy algorithm). This paper reviews the most frequently used active contours in a practical application, comparing weights, manually obtained by food technology experts, to volumes, automatically achieved by computer vision results. An experiment has been designed to recognize muscles from Magnetic Resonance (MR) images of Iberian ham at different maturation stages in order to calculate their volume change, using different active contour approaches. The sets of results are compared with the physical data. The main conclusions of the paper are the excellent correlation established between the data obtained with these three non-destructive techniques and the results achieved using the traditional destructive methodologies, as well as the real viability of the active contours to recognize muscles in MR images. © Springer-Verlag 2004.