.png){kind=icon}
.png){kind=icon}
Return
Efficient binary code indexing with pivot based locality sensitive clustering
.png){kind=icon}
PRE
AI
DOI：10.1007/s11042-012-1354-z.png)
.png){kind=icon}
Original request for help.png){kind=icon}
Share.png){kind=icon}
Save
Abstract
En 中文
High-dimensional indexing is fundamental in multimedia research field. Compact binary code indexing has achieved significant success in recent years for its effective approximation of high-dimensional data. However, most of existing binary code methods adopt linear scan to find near neighbors, which involve unnecessary computations and thus degrade search efficiency especially in large scale applications. To avoid searching codes that are not near neighbors with high probability, we propose a framework that index binary codes in clusters and only codes in relevant clusters are scanned. Consequently, Pivot Based Locality Sensitive Clustering (PLSC) is proposed and Density Adaptive Binary coding (DAB) method in PLSC clusters is presented. PLSC uses pivots to estimate similarities between data points and generates clusters based on the Locality Sensitive Hashing scheme. DAB adopts different binary code generation methods according to cluster densities. Experiments on open datasets show that offline indexing based on PLSC is efficient and DAB codes in PLSC clusters achieve significant improvement on search efficiency compared to the state of the art binary codes.
Keywords:
High dimensional indexing
Similarity search
Pivot based locality sensitive clustering
Density adaptive binary coding
Journal
IF:
3
Papers:
1.9W
Citations:
3.2W