基于客观聚类的手写数字识别方法  

A Handwriting Digital Recognition Method Based on Enhanced Objective Cluster Analysis

作　　者：王娜 胡超芳[4] WANG Na;HU Chaofang(Department of Automation, School of Electrical Engineering and Automation, Tianjin 300387, China;Tianjin Key Laboratory of Electrical and Electrical Technology, Tianjin Polytechnic University, Tianjin 300387, China;Tianjin Key Laboratory of Micro Optical Electronic Mechanical System Technology, Ministry of Education, Tianjin 300072, China;Department of Automation, School of Electrical and Information Engineering, Tianjin University, Tianjin 300072, China)

机构地区：[1]天津工业大学电气工程与自动化学院自动化系,天津300387 [2]天津市电工电能新技术重点实验室,天津300387 [3]天津大学微光机电系统技术教育部重点实验室,天津300072 [4]电气自动化与信息工程学院自动化系,天津300072

出　　处：《复杂系统与复杂性科学》2019年第2期77-84,94,共9页Complex Systems and Complexity Science

基　　金：天津大学微光机电系统技术教育部重点实验室开放课题基金(MOMST2016-4)

摘　　要：针对现有手写数字识别方法对噪声和图像结构敏感,易导致识别准确度下降,且计算过程复杂的问题,引入客观聚类算法并结合模板匹配机制,通过对待识别数字模板集的一次聚类以降低噪声和数据分布对聚类结果的影响,提高了识别结果的准确性;并利用新聚类中心约简原始模板数据集,实现计算效率的提高。通过对随机手写数字在结构变形和添加噪声等情况下仿真,并与传统手写数字识别方法比较,验证了所提方法的简单易行和有效性。The handwriting digital recognition methods generally are sensitive to the noise and the structure of image, which easily leads to the decrement of recognition accuracy and the increment of computation complexity. In this paper, the objective cluster analysis algorithm is introduced and combined with the template matching mechanism. In order to reduce the effects of the noise and the data distribution, and to improve the recognition accuracy, the one-pass clustering for the template set of the numeral to be identified is proposed. Furthermore, the new clustering centers are used to simplify the primary template dataset, by which the computation efficiency can be enhanced. The simulation about the random handwriting recognition in presence of structural deformation and noise demonstrates the simplicity, practicability and effectiveness of the proposed approach by comparing with the traditional methods.

关 键 词：手写数字识别 客观聚类 鲁棒 噪声 模板约简 

分 类 号：TP273[自动化与计算机技术—检测技术与自动化装置]