基于Spark平台的人脸图像检索系统  

Face Image Retrieval System Based on Spark Platform

作　　者：陈新荃 陈晓东[1] 蒋林华[3] 

机构地区：[1]中国科学院上海高等研究院,上海201210 [2]中国科学院大学,北京100049 [3]上海理工大学,上海200093

出　　处：《计算机工程》2018年第2期251-256,共6页Computer Engineering

摘　　要：传统人脸图像检索技术处理大规模图像数据时检索效率较低。为此,基于视觉词袋模型与Spark分布式计算平台构建人脸图像检索系统。根据人脸图像空间分布特点提出局部区块划分方法,减少视觉特征数并提高流程并行度,同时结合SURF局部特征和HOG区块特征设计候选图像相似得分算法,提高检索准确率。实验结果表明,与基于Hadoop的检索系统相比,该系统索引构建和检索的效率较高,并且在海量图像数据场景下具有良好的可扩展性和并发性。Traditional face image retrieval technology has lower retrieval efficiency when processing large-scale image data.Aiming at this problem,a face retrieval system based on Bag-of-Visual-Words（BoVW） model and Spark distributed platform is constructed in this paper.A local block partition method is proposed according to the spatial distribution of a face image,so as to reduce the number of visual features and enhance parallelism.By combining the Speed-up Robust Feature（SURF） local features and Histogram of Oriented Gradient（HOG） block features,a similarity algorithm of candidate images is designed to improve the retrieval accuracy.Experimental results show that the efficiency of the index construction and image retrieval in the proposed system are higher than those of the retrieval system based on Hadoop.The proposed system also has good scalability and concurrency under the massive image data scene.

关 键 词：人脸图像检索 分布式计算 区块匹配 相似度 视觉词袋模型 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]