Abstract | ||
---|---|---|
Many computer vision tasks rely on correspondence between features. SIFT and related descriptors represent features as high-dimensional vectors, which are used to compute a distance between features for matching. Calculating these distances is expensive in large datasets, so approximate nearest neighbour (ANN) approaches are used. ANN schemes that can be efficiently parallelised have been proposed, some of which divide up high-dimensional vectors into lower-dimensional subspaces. However, the feature descriptors computed by SIFT and similar methods are not computed in a homogeneous manner. There are clear statistical patterns within and between the components of feature vectors, and in this work we show that these patterns can have a strong effect on approximate nearest neighbor searching algorithms that are based on space subdivision. |
Year | DOI | Venue |
---|---|---|
2015 | 10.1109/IVCNZ.2015.7761548 | 2015 International Conference on Image and Vision Computing New Zealand (IVCNZ) |
Keywords | Field | DocType |
optimal space subdivision,parallel approximate nearest neighbour determination,computer vision,SIFT,high-dimensional vectors,ANN schemes,feature descriptors,statistical patterns,feature vectors,approximate nearest neighbor searching algorithms | k-nearest neighbors algorithm,Scale-invariant feature transform,Approximation algorithm,Feature vector,Search algorithm,Pattern recognition,Computer science,Linear subspace,Subdivision,Artificial intelligence,Cluster analysis | Conference |
ISSN | ISBN | Citations |
2151-2191 | 978-1-5090-0358-7 | 0 |
PageRank | References | Authors |
0.34 | 9 | 5 |
Name | Order | Citations | PageRank |
---|---|---|---|
huan feng | 1 | 45 | 4.43 |
Steven Mills | 2 | 41 | 17.74 |
David M. Eyers | 3 | 477 | 45.90 |
Xiaolong Shen | 4 | 2 | 2.07 |
Zhiyi Huang | 5 | 91 | 19.14 |