机构地区: 河南警察学院
出 处: 《现代电子技术》 2016年第9期66-69,共4页
摘 要: 针对现有手写数字识别难以处理几何变换下的识别难题,提出一种新的基于Grassmann流形度量的手写体数字识别方法。在分析不同几何变换下的手写数字字符所构成的非线性流形空间结构基础上,定义了Grassmann流形及其距离度量,并通过计算待识别数字字符与训练字符集合构成的Grassmann流形距离实现手写数字字符的分类识别。通过在MNIST数据库上的实验,证明该算法具有较好的实时性和鲁棒性,在对数字的识别率、稳定性、计算效率上显著优于现有基于切距离的手写数字识别算法,在识别率、稳定性上较现有基于欧氏距离的算法有较大的提高。 Since it is hard for the existing recognition method to recognize and handle the handwritten numbers in geometric transform,a new handwritten numeral recognition method based on Grassmann manifold measurement is proposed. On the basis of the analysis of the nonlinear manifold space structures composed of handwritten numeral characters in different geometric transforms,the Grassmann manifold and its distance measurement are defined. The waited recognition numeral characters are calculated and Grassmann manifold distance composed of character set is trained to realize the classification and recognition of the handwritten numeral characters. The results of experiment with MNIST database show that the proposed algorithm has better real-time performance and robustness,and is superior to the available handwritten numeral recognition algorithm based on tangent distance in the aspects of numeral recognition rate,stability and computing efficiency,and the recognition rate and stability of the proposed algorithm are better than those of the algorithm based on Euclidean distance.