The CMVAI-File: An Efficient Approximation-Based High-Dimensional Index Structure

Similarity-Search based index structure is an important topic in many application areas, such as content-based multimedia information retrieval, data mining, cluster analysis, etc. In order to improve query performance in high-dimensional vector spaces, lots of index structures have been proposed. However, many well-known index structures encounter “dimensional curse” while the dimension of feature vector increase. In order to solve such problem, researchers analyzed the nearest-neighbor search problem in high-dimensional vector spaces deeply and proposed some approximation-based index structures. The VA-File which is the first implement approximation-based index structures has done well in solving “dimensional curse” problem. However, approximate vectors in VA-File are just simply placed in flat structure files without other optimization methods. In this paper, we propose the Collecting and Multi-Vector-Approximation Indexed-File (CMVAI-File). The CMVAI-File is still an approximate-based index structure and uses a filter-based approach. Unlike the VA-File, the CMVAI-File uses multi-level approximate, collect the same vector-approximations and a segmental structure to improve query performance. Experimental results show that CMVAI-File has a promising improvement in performance.