I. Introduction
WORKS on the analysis of articular bone surfaces have mostly focused on local features. Yet, global characteristics accounting for kinematic functionalities, such as centers and axes of joints are also important to the orthopedists. Their location is estimated by fitting specific surface primitives: spheres at the hip and the shoulder, and circular cylinders at the knee (see references in [1]). Most methods still use mere least squares. We have more widely addressed the generic shape characterization based on global symmetries within all articular surfaces [1]. To this end, the use of a common quadratic model has been proposed. Its intention is to extract global features, while discarding topographic details. The estimation method must therefore be robust to tolerate the inherent local inaccuracy of such a simple model. Minimizing an objective function using robust techniques is usually most appropriate; a Hough-transform approach is impracticable at this parameter space dimension.