欢迎来到天天文库
浏览记录
ID:33367407
大小:940.44 KB
页数:53页
时间:2019-02-25
《聚类分析在图像分类中的应用研究》由会员上传分享,免费在线阅读,更多相关内容在学术论文-天天文库。
1、内蒙古科技大学硕士学位论文摘要到目前为止图像的分类识别依然是图像处理领域中的难点因为现实世界是多样的和复杂的获取图像的途径也是多种多样的这就使得不同的图像相互之间差别很大难以用统一的方法和模型描述论文中将数据挖掘中的聚类分析方法应用到图像分类中首先对基于密度和密度可达聚类算法ClusteringAlgorithmBasedonDensityandDensityreachable,CADD进行了深入分析并进行了大量的实验验证就CADD算法在大型图像数据集中应用暴露出来的不足做了四点改进(1)重写了计算并保存相异度矩阵的代码主要是为了降低保存相异度矩阵所占用
2、的内存空间(2)新程序引入了一个新的参数簇个数阈值(3)为了使原CADD算法能够更有效地处理变密度噪声和孤立点改进了原有密度可达距离的计算方法(4)新增了一种数据对象相似度度量的计算方法余弦相似度方法其次将改进了的CADD算法与传统的聚类算法K-means和层次聚类算法在真彩色BMP位图的分类中的实验结果作了对比分析得出的结论是(1)改进了的CADD算法与K-means和层次聚类算法相比具有较高聚类精度分辨率(2)尽管在聚类过程中CADD算法也需要输入初始参数密度参数σ和初始密度可达距离调节系数coefR但实验研究表明密度参数σ的变化对聚类结果影响不大根
3、据其定义初始密度可达距离调节系数coefR04、theimageisstilldifficultyintheimageprocessingfield.Becausetherealworldisdiverseandcomplex,thewayofobtainingimagesisalsovarious;whichmakesdifferentimagesvarygreatlywitheachotherandithardtouseuniformwaysormodelsdescribing.Theclassificationoftheimageusedtheclusteringanalysismethodin5、thedatamininginthispaper.Firstofall,wehavecarriedoutalargenumberofexperimentalverificationbasedonClusteringAlgorithmofDensityandDensityreachable(ClusteringAlgorithmBasedonDensityandDensityreachable,CADD),andimprovedfourdeficienciesofCADDalgorithmexposedintheapplicationoflarge-sca6、leimagedataaggregate.(1)Rewrotethecodeofthe"calculateandsavedifferentmatrix"toreducetheoccupiedmemoryofthepreservationofthedifferentmatrix.(2)Newproceduresintroducedanewparameter-thenumberofclusterthreshold.(3)Improvedthecalculatingmethodofthedensityuptodistance,inordertomaketheo7、riginalCADDalgorithmcaneffectivelyprocessthevaryingdensitynoiseandisolatedpoint.(4)Addedanewcalculatingmethodofdataobjectsimilaritymetric-cosinesimilaritymethods.Secondly,theexperimentalresultsoftheimprovedCADDalgorithmandthetraditionalK-meansclusteringandHierarchicalClusteralgor8、ithmintruecolorbitmapBMPclassificationwe
4、theimageisstilldifficultyintheimageprocessingfield.Becausetherealworldisdiverseandcomplex,thewayofobtainingimagesisalsovarious;whichmakesdifferentimagesvarygreatlywitheachotherandithardtouseuniformwaysormodelsdescribing.Theclassificationoftheimageusedtheclusteringanalysismethodin
5、thedatamininginthispaper.Firstofall,wehavecarriedoutalargenumberofexperimentalverificationbasedonClusteringAlgorithmofDensityandDensityreachable(ClusteringAlgorithmBasedonDensityandDensityreachable,CADD),andimprovedfourdeficienciesofCADDalgorithmexposedintheapplicationoflarge-sca
6、leimagedataaggregate.(1)Rewrotethecodeofthe"calculateandsavedifferentmatrix"toreducetheoccupiedmemoryofthepreservationofthedifferentmatrix.(2)Newproceduresintroducedanewparameter-thenumberofclusterthreshold.(3)Improvedthecalculatingmethodofthedensityuptodistance,inordertomaketheo
7、riginalCADDalgorithmcaneffectivelyprocessthevaryingdensitynoiseandisolatedpoint.(4)Addedanewcalculatingmethodofdataobjectsimilaritymetric-cosinesimilaritymethods.Secondly,theexperimentalresultsoftheimprovedCADDalgorithmandthetraditionalK-meansclusteringandHierarchicalClusteralgor
8、ithmintruecolorbitmapBMPclassificationwe
此文档下载收益归作者所有