Paper Info
Title
An efficient similar image search framework for large-scale data on cloud.
Abstract
In this paper, we aim to build a content-based image retrieval (CBIR) based on Map Reduce distributed processing framework, which is an image analysis platform with the capacities of operating stably and achieving the high performance in a large-scale dataset. It provides a fast and robust technique SURF to build feature descriptors for representing an image. As well, a building index scheme based on unsupervised learning method VP-Tree is also applied to construct a binary search tree for improving the efficiently while searching. Moreover, a locality distance cache is implemented on each processing machine, which maintains distance computations of previous query sessions for approximating the tightness distances between the current query object and objects stored in the database, thus decreasing needed time to process a single query. Especially, a Redis Master/Slave replication model is configured as a collaborative caching service for increasing data availability in the distributed environment. Finally, an MapReduce framework is exploited to retrieve the top-relevant images in searching phase. Regarding to experiments, it showed that our proposed framework achieves both accuracy and low responding time when working on a large dataset.
Year
DOI
Venue
2017
10.1145/3022227.3022291
IMCOM
Field
DocType
Citations 
Data mining,Locality,Distributed Computing Environment,Cache,Computer science,Image retrieval,Unsupervised learning,Artificial intelligence,Machine learning,Nearest neighbor search,Binary search tree,Cloud computing
Conference
0
PageRank 
References 
Authors
0.34
8
2
Name
Order
Citations
PageRank
Tri-Dung Nguyen1299.02
Eui-Nam Huh210011.43