대용량 비디오 데이터베이스를 위한 인덱싱

초록

Multimedia databases require an indexing technique that is able to provide high performance with high dimensional data. Unlike other tree based indexing techniques, LSH inherits the characteristics of hashing, and thus is suitable for high dimensional data. However, it is difficult to apply LSH to real life databases because the performance of LSH is quite sensitive to its parameters and also the properties of datasets. In this research, we first analyze the characteristics of LSH and the changes of its accuracy and performance according to different parameter settings through extensive experiments in a video database environment. In addition, based on the results and analyses, we propose four new hash functions that reflect the properties of high dimensional datasets and show their effectiveness in terms of accuracy and performance via a series of experiments.

키워드