Approximate nearest neighbour (ANN) search has been favourable for large-scale information retrieval in the recent past, and many hashing techniques for ANN have been proposed for retrieval of data in a large database, given a query. Hashing based indexing techniques are being mostly favoured for similarity search from huge database because of its efficiency in retrieval accuracy and low memory requirements. The long code length of randomised hashing based indexing techniques gives good precision but required more computational cost and high memory cost. DSH uses K-means algorithm to partition n data points into k groups for quantisation of data. This paper addresses the problem of long hash codes, computational cost, long convergent time and high memory requirements to achieve efficient similarity searching. Experiment was setup and Geo-SPEBH was evaluated on SIFT 1B based on MAP, precision-recall metrics and GeoSPEBH outperformed the state-of-the-art techniques.
展开▼
