Improved VA-file For Approximate Image Retrieval

摘要：

For similarity search in high-dimensional vector spaces,the conventional index structures performance degrades as the number of dimensions increases. VAfile has been proved to be the only method that can improve search efficiency by far. This method based on approximations to make the unavoidable sequential scan as fast as possible and reduce I/O time. The efficiency of VA-file is mostly depended on the filtration efficiency in the filtering step. In this paper,we improve the filtration efficiency by changing the way of choosing split position. We also improve VA-file by saving middle caculate result between original vector and compressed vector,which achieves good result especially when database is large.

关键词： VA-file approximate similarity search highdimensional indexing curse of dimensionality

作者: Li Yuqin Zhang Song Lv Xueqiang Shi Shuicai

作者单位: Beijing Information Science and Technology University,Chinese Information Processing Research Center Beijing Information Science and Technology University,Chinese Information Processing Research Center

会议类型: 国际会议

会议名称: 2010 International Forum on Computer Science-Technology and Applications(2010 国际计算机科学技术应用论坛 IFCSTA 2010)

会议地点: 南宁

会议语种:英文

页码: 54-56

在线出版日期: 2010-12-10（万方平台首次上网日期，不代表论文的发表时间）