Improvement of Feature Point Annotation for Three-Dimensional Spine Model Based on Curvature Multiple Feature Fusion

HUI Yu,WU Jun-sheng,YU Bin,ZHANG Chen,WU Zhen

DOI: https://doi.org/10.3788/ope.20162411.2872

2016-01-01

Abstract:Because the precise coordinate description and accurate matching of physiology of the lumber spine have not yet reached the accuracy requirement of the medicine ,this paper researches how to describe the physical coordinates of lumber spine feature points precisely . The definition and traditional annotation method of human spine sample features were introduce . Then ,an improved method to adaptively annotate the feature points was proposed based on curvature multiple feature fusing to solve the problem of weaker precision for the manual annotation of feature points .In this method ,the definiting values of Gaussian curvature and average curvature of a certain feature point were found out firstly ,and a relative maximum value of curvature of the feature point was obtained . Then ,the relative maximum values of all the model points within a specified minimum radius r were obtained .Since the larger the curvature of maximum value was ,the larger the crook degree of the 3-dimensioanl model surface on the point was ,and the point can show the outline of the 3-dimensional model .Therefore ,the maximum value was taken as the curvature description of the feature point to replace the manual pick up point ,thus the feature variation of the point could be described precisely . Finally ,the verification results were analyzed ,and the results indicate that the improved method increases the accuracy by 37% as compared with that of traditional annotation methods .These results demonstrate that the new method is feasibility and effectiveness .