General bases indexing using approximate structure techniques

The emergence of numerical technologies requires the use of powerful tools and retrieval engines for fast and efficient access to images datasets. In spite of the rapid growth of computing performance, it is always difficult to manage huge amount of data because of the exponential growth of the processing time according to the data complexity. Therefore, in this paper, Approximate Nearest-Neighbor (ANN) algorithms are used as a solution of dramatically improving the retrieval speed. Indeed, we focus on locality-sensitive hashing (LSH) technique. Since its performance depends essentially on the hash function used to partition the space, we propose to introduce a new function inspired from the E₈ lattice and to combine it with the Multi-Probe-LSH and the Query Adaptative LSH (QA-LSH). This method is applied in our case in the context of CBIR. In order to prove the robustness of the proposed approach, a set of experimental results are compared with similar state of the art algorithms.