Optimal initialization for 3D correspondence optimization: an evaluation study

摘要: The identification of corresponding landmarks across a set training shapes is prerequisite for statistical shape model (SSM) construction. We automatically establish 3D correspondence using one new and several known alternative approaches consistent, shape-preserving, spherical parameterization. initial determined by all employed methods refined optimizing groupwise objective function. quality models before after optimization thoroughly evaluated data sets clinically relevant, anatomical objects varying complexity. Correspondence benchmarked in terms the SSMs' specificity generalization ability, which are measured different surface based distance functions. We find that our approach performs best complex objects. Furthermore, previously published own allow (i) building SSMs significantly better than well-known SPHARM method, (ii) establishing quasi-optimal low moderately without additional optimization, (iii) considerably speeding up convergence, thus, providing means practical, fast, accurate SSM