柔性体点分布模型的时变插值算法

Computing a middle model from multiple flexible models

相对于刚体,柔性体的建模和分析比较复杂,点分布模型是目前最有效的柔性体建模方法之一。然而,在许多条件下,需要依据研究对象的时间变化特性构建多个即时模型或者任意时刻的点分布模型。针对这种情况提出一种插值算法,根据两个相邻的已知点分布模型生成某个中间时刻的过渡模型。以线性插值为例,可以在任意给定时刻生成一个插入模型的平均形状和特征表达。其中,平均形状可由前后相邻模型的线性组合计算得到,而特征表达则需要根据相应协方差矩阵的特征向量分析得到。经过了计算机仿真和实际图像验证,结果表明,这种模型插值技术具有良好的可行性和实用价值。

The point distribution model is a powerful shape description technique and has enormous applications in a short time, especially for flexible or deformable model analysis in images. Initially, a model is generatively built by statistical analysis from a number of examples. In some applications, nevertheless, it allows building a set of models or continuous sequence of models since a single one is not well representing a highly changing object. This paper presents a method to compute such a middle time-point model which is formed from two neighboring models in runtime. In case of linear algorithm, the model parameters that have already given by PCA are determined in several steps. The methodology is validated with both computer simulation and practical experiments. Results indicate that it is worthwhile for its adoption in some complex modeling applications.