基于信息熵加权的三维模型特征融合算法

A 3D Model Feature Fusion Algorithm Based on Entropy Weights

特征融合是提高三维模型检索有效性的一种重要手段,越来越受到广泛关注.为解决融合特征权重的自适应问题,提出了一种基于信息熵加权的三维模型多特征融合算法,以提升检索效果.算法的主要贡献有:1)为了精确地刻画形状分布直方图,针对传统形状分布算法提出了三次样条插值形状分布特征提取算法,同时由于形状分布算法本身缺少描述模型形状的层次分布信息,设计了简化的形状上下文特征提取算法;2)采用衡量不确定性的信息熵来自适应计算上述两个欲融合特征的权值,进而得到融合后的相似距离用于三维模型检索.在SHREC2014的综合模型库上实验,结果表明:所提特征融合算法的通用和加权的6种评价指标(PR/NN/FT/ST/E/DCG)优于采用单一特征的检索算法,且运行效率较高.

Feature fusion is an important tool to improve the retrieval effectiveness of 3D models,which is being concerned widely.To target at advancing the retrieval results and solving the problem of automatically adapting the weight for each fusion feature,a novel multi-feature fusion algorithm of3 D model based on entropy weights(MFF-EW)is proposed.The main contribution of the algorithm is listed as follows.Firstly,cubic spline interpolation-shape distribution algorithm(CSI-SD)is put forward to characterize shape distribution histogram more precisely on the basis of traditional shape distribution algorithm; meanwhile,simplified shape context algorithm(SSC)is designed to compensate for the hierarchy distribution information of a model,which lacks in CSI-SD algorithm.Afterwards,the information entropy,a method to measure uncertainty,is used to adaptively calculate the weights of two fusion features mentioned above,and then the fusion similarity distance for 3D model retrieval can be obtained.The large-scale comprehensive benchmark in SHREC2014 is used as testing. Experimental results show that the proposed MFF-EW algorithm outperforms the corresponding algorithms according to the general and weighted six evaluation metrics(PR/NN/FT/ST/E/DCG).And the algorithm is also time-efficient.