Hadoop环境下基于PLDA的图像场景分类方法研究  

Image Scene Classification Based on PLDA in Hadoop

作　　者：李勋[1] 高洁琼[1] 杨景玉[1] 王松[1] 

机构地区：[1]兰州交通大学电子与信息工程学院,甘肃兰州730070

出　　处：《兰州交通大学学报》2015年第6期81-86,共6页Journal of Lanzhou Jiaotong University

摘　　要：针对大规模图像场景分类速度缓慢的问题,提出了一种Hadoop环境下基于PLDA的图像场景分类方法.首先利用Hadoop分布式存储系统HDFS存储图像数据;其次通过扩展MapReduce数据类型以及输入类型使得每幅图像作为一个MapReduce任务的数据处理单元;然后利用MapReduce分布式提取图像SIFT特征,并使用Bag-of-Features模型将其转化成词频向量;最后使用MapReduce实现了PLDA算法,并在此基础上完成了Hadoop环境下基于PLDA的图像场景分类.实验结果表明,该方法能够充分利用MapReduce的分布式特性,在分类结果合理的情况下,大幅提高了图像场景的分类速度.Because the speed of large scale image scene classification is slow, an image scene classi- fication method based on PLDA in Hadoop is proposed. Firstly, HDFS is used to store image da- ta. Secondly,MapReduce data type and input type are extended to make each image as a data slice which is not segmented. MapReduce and Bag-of-Features model are used to extract image SIFT features and to transform them into frequency vectors. Finally,MapReduce is used to implement PLDA algorithm. On this basis,image scene classification is finished based on PLDA in Hadoop. The results show that this method can make full use of the distributed characteristics of MapRe- duce and improve the speed of image scene classification greatly.

关 键 词：图像场景分类 PLDA HADOOP MAPREDUCE SIFT BOF 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]