Approximate Nearest Neighbor Search Using Enhanced Accumulative Quantization

Approximate nearest neighbor (ANN) search is fundamental for fast content-based image retrieval. While vector quantization one key to performing an effective ANN search, in order further improve accuracy, we propose enhanced accumulative (E-AQ). Based on our former work, introduced the idea of quarter point into (AQ). Instead finding centroid, was used quantize and computed each according its centroid second centroid. Then, error produced through codebook training reduced without increasing number centroids codebook. To evaluate accuracy which vectors were approximated by their outputs, realized E-AQ-based exhaustive method search. Experimental results show that approach gained up 0.996 0.776 Recall@100 with eight size 256 codebooks SIFT GIST datasets, respectively, at least 1.6% 4.9% higher than six other state-of-the-art methods. Moreover, based experimental results, E-AQ needs fewer while still providing same accuracy.