基于Isomap的手写体数聚类研究.docVIP

基于ISOMAP的手写体数字聚类研究 摘要：手写数字聚类技术在手写数字识别、触控操作、图像处理等方面都有重要的理论意义和实用价值。目前，这方面的最新研究成果主要有细化算法，LLE和SVM算法，NLDR方法。 本文简要阐明了基于流形学习的手写数字聚类的目的、意义及发展现状，详细论述了手写数字识别的原理及实现过程，系统介绍了目前主要的PCA、Iosmap等降维算法。介绍了两种聚类算法——Kmeans及模糊C均值。利用该算法的思想及原理，对数字聚类进行研究并进行仿真实验，计算不同聚类算法的错误率并对比各种算法的优缺点，找到最优化的算法，提出自己的改进方法。实验提取了10组数据，每组包含300个数据值，聚类的错误率达40%以上，后经过改进降维参数，增加提取的数据量，成功的降低了错误率。另外又变换降维方法和聚类方法，与前一组实验做对比，得出ISOMAP的聚类效果较理想，同时发现聚类效果受数据量大小的影响呈正弦曲线。 在文章结尾处，概述数字识别技术在未来生活及科技方面的广阔应用前景，最后对全文作出总结，对数字识别及聚类技术在这方面存在的问题和进一步的研究方向进行展望。 关键词：流形学习 数字识别 Iosmap Kmeans 模糊C均值 Research on handwritten numeral clustering Based on isometric feature mapping ( ISOMAP) Abstract：The technology of handwritten numbers cluster have an important theory significance and practical merit in handwritten numeral recognition, touch operation, image processing etc.At present,the latest research in this discipline are thinning algorithm,LLE and SVM ,NLDR.But they have some prolems in manifold discontinuity dimensionality reduction,noise analysis,estimate the intrinsic dimension of the original data and so on are not being addressed property. This paper briefly expounds the purpose of handwritten digital cluster based on manifold learning and significance and the present development situation.the paper discusses in detial the handwritten digital recognition principle and realization process.it also introduces the method of LLE,PCA,ISOMAP etc.and k-means,fuzzy c mean.Using the idea and principle of the algorithm, to carry on the research of digital cluster and simulation pute the error rate and compare the advantages and disadvantages of different clustering algorithms,find the best method and give out my option. At the end of article,overview the broad application prospects of handwritten numeral clustering in various field of science. prospect of the digital recognition technology in the future. Keywords: manifold learning numerical recognition LLE Iosmap Kmeans 目录 中文摘要……………………………………………………………………………1 英文摘要………………………………………………………………