基于高斯分布的强独特性描述子采样模型研究

Research on Sampling Model of Strong Uniqueness Descriptor Based on Gaussian Distribution

针对基于传统采样模型的特征描述算法鲁棒性不高的问题,建立一种基于高斯分布的强独特性描述子采样模型.首先通过对比BRISK和FREAK采样模型,确定了影响描述子性能的模型参数;然后通过理论建模分析参数对模型性能的影响规律,选取最优值参数以达到信息含量最优、独特性最强的目的;最后根据人眼视网膜细胞的分布特性构建改进模型.实验结果表明,基于改进模型的描述算法不仅可以更好地克服图像中各种尺度、旋转、视角和噪声等变换的影响,而且完全满足实时性要求,其鲁棒性与BRISK和FREAK采样模型相比分别提升9%和5%.

The binary descriptor based on traditional sampling model has low robustness performance.To solve this serious problem,a sampling model of strong uniqueness descriptor based on Gaussian distribution is proposed.Based upon the comparison of BRISK and FREAK sampling models,the parameters are determined which affect the model performance.Then in order to make sure of the algorithm＇s information content and uniqueness,the effect of parameters is analyzed to choose optimal ones by theoretical modeling.Finally the model is gained with the combination of biotical principle.Comparative experiments show that the descriptor applying the proposed sampling model can not only better overcome the influence of various change of scale,viewpoint and noise in the images,but fully meet the real-time requirements,and the robust performance has improved by 9% and 5% respectively compared with the BRISK and FREAK models.