摘要
A novel idea,called the optimal shape subspace （OSS） is first proposed for optimizing active shape model （ASM） search.It is constructed from the principal shape subspace and the principal shape variance subspace.It allows the reconstructed shape to vary more than that reconstructed in the standard ASM shape space,hence it is more expressive in representing shapes in real life.Then a cost function is developed,based on a study on the search process.An optimal searching method using the feedback information provided by the evaluation cost is proposed to improve the performance of ASM alignment.Experimental results show that the proposed OSS can offer the maximum shape variation with reserving the principal information and a unique local optimal shape is acquired after optimal searching.The combination of OSS and optimal searching can improve the ASM performance greatly.
首先提出了最优形状子空间概念,它由主要形状子空间和主要形状变化子空间联合构成,最大程度上包含了搜索形状上的变化,更贴近现实中要搜索的目标.然后,在仔细研究了经典算法中的搜索过程后,通过引入代价函数和反馈机制,提出了一种最优搜索的概念,使在搜索过程中搜索、评价、反馈不断地进行,最后得到最佳的搜索结果.实验表明:提出的最优形状子空间在保证主要形状的基础上给出了最大限度的形状变化,最优搜索过程可保证搜索到局部的惟一最优形状.它们的综合大大改善了动态形状模型的性能,并提高了搜索的精确性.
基金
21st Century Education Revitalization Project (No.301703201).
