Surface shape and curvature scales
Image and Vision Computing
Automatic Detection of the Back Valley on Scoliotic Trunk Using Polygonal Surface Curvature
ICIAR '08 Proceedings of the 5th international conference on Image Analysis and Recognition
A pose-independent method for 3D face landmark formalization
Computer Methods and Programs in Biomedicine
| Hi-index | 0.00 |
Anatomical landmarks Biomechanics Curvature maps The landmarks on the body surface are important to shape and motion analysis. It is much better if the landmarks are anatomical ones, which are independent of position and coordinate system. The objective of this method is to present an easy-implemented method for extracting anatomical landmarks on the cylindraceous body surface which could be used in motion analysis or in medical treatment. The surface is scanned by FastSCAN (Polhemus, Colchester, Vermont, USA) and described by scattered three-dimensional surface points. The method provides the estimation of second-order derivatives by way of least-squares surface fitting to calculate the Gaussian curvature and mean curvature. To separate convexity from concavity, the Koenderink shape index maps of foot and leg are given as examples. The landmarks formed by underlying muscles and skeletal structures such as the malleoli distinguish themselves clearly on the Koenderink shape index maps. Minutes after the foot and leg are scanned, the curvature maps of the foot and leg provide the shape information and the loci of landmarks avail the statistical shape analysis as well as foot underside deformation analysis. Furthermore, the anatomical landmarks around the knee and ankle, defining the transcondylar and transmalleolar axis, make it possible to calculate the tibial torsion by this non-invasive way. Generally, this method is fast and accurate. However, it gives some inaccurate results on the patch edge, which should be interpreted with caution, when it is applied on a surface patch in other occasions because of a small number of points unevenly distributed in the operator. Presently, the method is computationally intensive although the time can be reduced to a few seconds at the sacrifice of image resolution. Further efforts will be made to get the real-time information.