基于Hadoop的视觉词袋模型图像分类算法  

Image Classification Approach of Bag of Visual Words Model Based on Hadoop

作　　者：侯春萍[1] 张倩楠[1] 王宝亮[2] 常鹏[2] 孙韶伟[2] 

机构地区：[1]天津大学电气自动化与信息工程学院,天津300072 [2]天津大学信息与网络中心,天津300072

出　　处：《天津大学学报（自然科学与工程技术版）》2017年第6期643-648,共6页Journal of Tianjin University：Science and Technology

基　　金：国家自然科学基金资助项目(61571325)~~

摘　　要：随着互联网的发展和数字图像获取技术的进步,传统图像分类算法在处理海量数字图像时,面临耗时过多、文件系统及处理架构落后的问题.针对这一问题,利用主流的Hadoop开源分布式计算平台,引入视觉词袋模型实现对图像的表示,并对模型的图像直方图化过程做出改进,提出一种自适应的特征分配方法,最后采用易于并行的随机森林算法作为分类器,以充分利用Hadoop平台强大的分布式计算能力.实验显示,基于Hadoop平台的图像分类方法在处理大规模数据集时较单机环境能有效减少时间消耗,同时具有良好的分类效果.As the Internet grows and technology of acquiring digital images advances rapidly,problems with the conventional image classification methods gradually arise while dealing with massive digital images,such as being time-consuming and lacking timely update of the file system and processing architecture. To combat this problem,an image classification approach is proposed based on Apache Hadoop,the mainstream open-source distributed process-ing system. Firstly,the bag of visual words（BoVW）model was utilized to achieve simplified image representa-tions. Meanwhile,an improvement was made to the model during the histogram representation period and an adap-tive soft assignment algorithm was proposed. Lastly,the easy-paralleled random forest algorithm was employed as the classifier so as to make full use of the advantages of the platform. Experiments show that the proposed method of image classification based on Hadoop could effectively decrease the computing time compared with single-PC method while dealing with mass images,and at the same time gain good classification results.

关 键 词：HADOOP 图像分类 视觉词袋 随机森林 软分配 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]