Adaptive reconstruction of anatomical surfaces from human body measurements

Abstract

A technique that adaptively fits a deformable Bezier surface to partial or whole human body measurements for free-form surface reconstruction is described. The proposed method utilizes an unconventional neural network, called a Bernstein basis function (BBF) network, which performs a weighted summation of rational Bernstein polynomial basis functions. Modifying the number of basis neurons is equivalent to changing the degree of the Bernstein polynomials. Each BBF network determines the control points of a low-order rational Bezier surface that best approximates the shape of randomly organized coordinate data. A key feature of the algorithm is that the measured data does not have to be re-ordered or parameterized prior to fitting. In addition, the rational Bezier surface retains the relative position of the parametric coordinates (u,v) as it deforms. Once the adaptation phase is complete, the weights of the network can be entered directly into a variety of commercially available geometric modeling and CAD/CAM packages for shape reconstruction. An experimental study is presented to demonstrate the effectiveness of the BBF network for generating rational Bezier surfaces.