科研之友
微信
新浪微博
Facebook
分享链接
摘要
In order to allow content-based retrieval of multimedia documents spread over large networks, we propose an indexing system based on decentralized and asynchronous learning of visual codebooks. We propose a Gossip-based decentralized algorithm to compute an accurate visual codebook at each networking node. We provide an empirical law to define the optimal parameters, given the targetted network size, to get codebooks which are equal between nodes with low communication costs. An experimental study highlights the scaling abilities and the retrieval accuracy of our system.
- 出版日期2015-3
- 单位INRIA
