基于Hadoop的人脸识别并行化方法的研究

Research on a Parallelized Method of Face Recognition Based on Hadoop

针对传统的人脸识别技术具有面向小数据量、单机处理环境、实时性低等问题,提出将目前流行的大数据处理框架Hadoop与人脸识别技术相结合的处理方法。将人脸数据库存放在本地,将人脸数据库中的人脸图像的URL组成的文本文件存放至HDFS,并且作为MapReduce的输入。经过实验对比,在较大数据量的时候,本文提出的人脸识别并行化方法的效率较单机环境下有明显提升,且集群表现出良好的稳定性和伸缩性,为人脸识别并行化提供了一种可行的方案。

In view of the traditional face recognition technology,which is oriented to small data,single machine processing environment,low real-time and so on,this paper proposes a processing method combining the current popular big data processing framework Hadoop with face recognition technology.The face database is stored locally,and the text file composed of the URL of the face image in the face database is stored to HDFS as an input to MapReduce.The efficiency of face recognition parallelization method proposed in this paper is obviously improved when the data is large,and the cluster shows good stability and scalability,which provides a feasible scheme for face recognition parallelization.