No abstract

No abstract

Traditional CBIR (Content Based Image Retrieval) system on parallel platform spends too much time facing largescale images. In this paper, we present an improved CBIR system based on Apache Spark, a fast and general large-scale data processing engine. At first, we do preliminary processing to build system environment and extract feature vectors. Then, in order to solve the problem of huge numbers of small files, we construct image database by using Apache Avro to combine these small files and improve structure of the common Avro file and so making it more suitable for our system. Finally, at the matching stage, we make the Avro file to be persisted in memory and reused conveniently. Multiple experimental results show our approach performs well in terms of both speed and accuracy rate.

No abstract

No abstract