Accelerating content-based image retrieval via GPU-adaptive index structure.

摘要: A tremendous amount of work has been conducted in content-based image retrieval (CBIR) on designing effective index structure to accelerate the process. Most them improve efficiency via complex structures, and few take into account parallel implementation underlying hardware, making existing structures suffer from low-degree parallelism. In this paper, a novel graphics processing unit (GPU) adaptive structure, termed as plane semantic ball (PSB), is proposed simultaneously reduce process exploit acceleration hardware. PSB, semantics are embedded generation representative pivots multiple balls selected cover more informative reference features. With online CBIR factorized independent components that implemented GPU efficiently. Comparative experiments with GPU-based brute force approach demonstrate can achieve high speedup little information loss. Furthermore, PSB compared state-of-the-art approach, random (RBC), two standard datasets, Corel 10 K GIST 1 M. Experimental results show our achieves higher than RBC same accuracy level.