Abstract
GEO-CYCLICAL STRUCTURE-BASED HASHING
Abubakar Usman Othman*, Boukari Souley, Abdulsalam Ya’u Gital and Iliya Musa Adamu
ABSTRACT
in the past few years, approximate nearest neighbor (ANN) search has become the core in search field such as machine learning, pattern recognition, and data mining were the applications involves huge data bases. Hashing-based ANN have been known for their low memory requirements and computational cost. For this, many hashing-based approaches have been proposed for efficient similarity search from large-scale image collection for a given query. However, most of these methods generates their projections randomly which increases the cost of storage, they do not also consider the search time, and the data points are not spatially coherently mapped into compacts binary codes. To address these draw back, this paper proposes a scalable algorithm named Geo-cyclical structure-based hashing that can explore and preserve the underlying geometric structure information of the data. Extensive experiment would show that the new scheme will be more robust and more effective in terms of performance compared with state-of-the-art hashing approaches.
[Full Text Article] [Download Certificate]