Curve and boundaries measurement using B-splines and virtual images

Abstract

This paper presents a digital image correlation method able to detect and measure a contour or silhouette in a digital image, with a sub-pixel precision. A virtual image that describes a simple contour or silhouette roughly similar to the physical object to detect is created. The matching with the physical image of the object is then performed based on digital image correlation. We look for the displacement field which transforms the physical image which is the closest to the virtual one. In order to keep a problem of finite dimension, this field is decomposed on a B-spline basis. The algorithm is close to the digital image correlation method; an efficient implementation is proposed that makes use of levelsets. The full method is tested over both synthesized and real images, one of a star shaped toy and one on an ill-defined image of a high temperature metallic specimen on which no other boundary measurement was possible. The influence of the arbitrary choice of the virtual image contour or silhouette arbitrary parameters is discussed with respect to the precision of the method. This one is shown to be sub-pixel in any case.