Estimating simple closed contours in images

Abstract

A technique for estimating the boundary of objects that may be described by a simple closed contour is presented. The problem is posed as one of maximum a posteriori (MAP) optimization. The method employs an elliptical Fourier series to describe the contour. A scaled derivative-of-Gaussian description of the object boundary cross-section is employed, coupled with guided gradient descent estimation of the contour coefficients. The technique is applied to X-ray images of the left ventricle, but is easily extensible to a wide variety of images of similar objects. It is shown that a few elliptical harmonics accurately model the ventricle outline. Accurate, robust estimation of the left ventricle outline from a single image is shown for a set of images.<>