基于外存的位置敏感搜索方法

摘要

Locality Sensitive Hashing is widely used in informational retrieval, object recognition and video visual search recently. Though they partly solved high dimension retrieval problem compared with tree-based method, their performance was limited for data structure stored in main memory. A locality sensitive search method based on extend memory is proposed on the basis of E2LSH, according to the experiments, the retrieval time of the method is decreased largely whereas the re-trieval accuracy is nearly at the same level with linear search.%位置敏感哈希在信息检索、目标识别和视频语义搜索等领域得到了广泛应用,与基于树的方法相比,它们虽然初步解决了高维检索问题,但这些基于主存的方法在实际应用中仍有较大的局限性.为解决大数据集快速检索问题,在E2LSH基础上提出了基于外存的位置敏感搜索方法,将数据集各点通过位置敏感哈希函数族进行映射并在外存建立索引文件,实验证明该方法在检索准确率几乎相当的情况下检索时间大大缩短.