DocumentCode :
2657625
Title :
Reduce Pruning Cost to Accelerate Multimedia kNN Search over MBRs Based Index Structures
Author :
Li, Hui ; Zhang, Xiao ; Wang, Shan
Author_Institution :
Key Lab. of Data Eng. & Knowledge Eng., Renmin Univ. of China, Beijing, China
fYear :
2011
fDate :
4-6 Nov. 2011
Firstpage :
55
Lastpage :
59
Abstract :
MBR (Minimum Bounding Rectangle) has been widely used to represent multimedia data objects for multimedia indexing techniques. In kNN search, MINDIST and MINMAXDIST was the most popular pruning metrics employed by MBR based index structures. However, with the increase of dimensionality, the computation costs of above two metrics are expensive, and the involved distance computations may become the considerable component. In this paper, to improve the performance of kNN search over multimedia data, we proposed an approach to reduce the computation cost of pruning heuristics by use the cheap upper bound of MINMAXDIST instead of its precise value, and then we derive enhanced cheap heuristics based on this upper bound for kNN pruning, which are helpful to avoid visiting unnecessary data objects and MBRs. We conducted an extensive experimental study on a large real-life multimedia dataset called MSRA-MM 2.0, obtained from the commercial Bing Image search engine. The results show that the proposed approach significantly reduced the computation cost and boosted the overall performance of MBR based multimedia kNN search.
Keywords :
data structures; database indexing; image retrieval; multimedia databases; pattern clustering; query processing; search engines; MBR based index structures; MINDIST; MINMAXDIST; MSRA-MM 2.0; cheap upper bound; commercial Bing image search engine; enhanced cheap heuristics; minimum bounding rectangle; multimedia data objects; multimedia indexing techniques; multimedia kNN search; pruning cost reduction; real-life multimedia dataset; Indexing; Measurement; Multimedia communication; Search problems; Streaming media; Upper bound; MBR; kNN search; multidimensional pruning; multimedia indexing;
fLanguage :
English
Publisher :
ieee
Conference_Titel :
Multimedia Information Networking and Security (MINES), 2011 Third International Conference on
Conference_Location :
Shanghai
Print_ISBN :
978-1-4577-1795-6
Type :
conf
DOI :
10.1109/MINES.2011.85
Filename :
6103721
Link To Document :
بازگشت