Reconstruction error of 3D anatomic structures represented by 2D Fourier Descriptors using a dissimilarity measure

The problem we address here concerns the quantification of the reconstruction error of 3D objects formed by two-dimensional (2D) cross-sectional adjacent slices. We introduce two distances (a shape distance and a size distance) in the descriptors space to compare contours at the same slice level and to perform a local error analysis. Then, a dissimilarity measure between two versions of the same 3D shape is used to compute a global error. The complete processing is performed without registration because of the geometrical invariancy property of the descriptors. Examples concerning the ulnar bone structure are given.