Metric Indexing for the Earth Mover's Distance
Article No.: 3, Pages 1 - 8
Abstract
The Earth Mover's Distance (EMD) has become a popular choice for applications in similarity search, particularly in applications such as few-shot image classification where it is observed to match human perceptions of image differences better than other distance measures such as the Euclidean distance. Currently, in domains such as few-shot image classification, it is common to use exhaustive search during query time which is inefficient in applications using distances with high computational complexity such as the EMD. Since the space in these applications is not guaranteed to be a vector space, existing techniques such as product quantization cannot be applied. In this paper, we study the application of metric space indexing structures towards the reduction of the number of EMD computations needed during query time. We conduct experiments on several image datasets in order to identify the advantages and disadvantages of different metric space data structures. These experiments have not been performed in the context of the EMD before and demonstrate that the VP-tree is more robust to increases in dataset complexity in this domain than comparable metric indexing data structures. Furthermore, we combine these data structures with deep feature extraction to develop a method for efficient deep image retrieval in metric spaces. Taking inspiration from distance stretching methods in the previous literature, we develop a novel approximate nearest neighbor algorithm for k-NN search that can greatly reduce the number of distance computations needed for retrieval without significantly changing k-NN accuracy.
References
[1]
A. Amir, A. Efrat, P. Indyk, and H. Samet. 1999. Efficient regular data structures and algorithms for location and proximity problems. In Proceedings of the 40th IEEE Annual Symposium on Foundations of Computer Science. New York, 160--170.
[2]
C-H Ang, H. Samet, and C. A. Shaffer. 1990. A new region expansion for quadtrees. IEEE Trans. on Pattern Analysis and Machine Intelligence 12, 7, 682--686.
[3]
M. Arjovsky, S. Chintala, and L. Bottou. 2017. Wasserstein gan. arXiv preprint arXiv:1701.07875.
[4]
I. Assent, A. Wenning, and T. Seidl. 2006. Approximation Techniques for Indexing the Earth Mover's Distance in Multimedia Databases. In Proc. of the 22nd Intl Conf. on Data Engineering, ICDE, Atlanta, GA, USA. IEEE Computer Society, 11.
[5]
M. Aumüller, E. Bernhardsson, and A. Faithfull. 2020. ANN-Benchmarks: A benchmarking tool for approximate nearest neighbor algorithms. Information Systems 87, 101374.
[6]
S. Berchtold, D. A. Keim, and H.-P. Kriegel. 1996. The X-tree: an index structure for high-dimensional data. In Proc. of the 22nd Intl Conf. on Very Large Data Bases (VLDB). Mumbai (Bombay), India, 28--39.
[7]
A. Beygelzimer, S. Kakade, and J. Langford. 2006. Cover trees for nearest neighbor. Proc. of the 23rd Intl Conf. on Machine learning.
[8]
L. Boytsov and B. Naidan. 2013. Engineering Efficient and Effective Non-metric Space Library. In Similarity Search and Applications - 6th Intl Conf., SISAP, A Coruña, Spain, Proceedings. Springer.
[9]
G-I Brokos, P. Malakasiotis, and I. Androutsopoulos. 2016. Using centroids of word embeddings and word mover's distance for biomedical document retrieval in question answering. arXiv preprint arXiv:1608.03905.
[10]
K. Grauman and T. Darrell. 2004. Fast contour matching using approximate earth mover's distance. Proc. of the IEEE Computer Society Conf. on Computer Vision and Pattern Recognition, I--I.
[11]
A. Guttman. 1984. R-trees: a dynamic index structure for spatial searching. In Proc. of the ACM SIGMOD Conf. Boston, 47--57.
[12]
K. He, X. Zhang, S. Ren, and J. Sun. 2016. Deep Residual Learning for Image Recognition. In IEEE Conf. on Computer Vision and Pattern Recognition (CVPR). 770--778.
[13]
G. R. Hjaltason and H. Samet. 2000. Incremental similarity search in multimedia databases. Computer Science Technical Report TR-4199. University of Maryland, College Park, MD.
[14]
G. R. Hjaltason and H. Samet. 2003. Properties of embedding methods for similarity searching in metric spaces. 25, 5, 530--549.
[15]
G. Huang, C. Guo, M. J Kusner, Y. Sun, F. Sha, and K. Q Weinberger. 2016. Supervised word mover's distance. In Advances in Neural Information Processing Systems. 4862--4870.
[16]
H. Jegou, M. Douze, and C. Schmid. 2010. Product quantization for nearest neighbor search. IEEE transactions on pattern analysis and machine intelligence 33, 1, 117--128.
[17]
J. Johnson, M. Douze, and H. Jégou. 2017. Billion-scale similarity search with GPUs. arXiv preprint arXiv:1702.08734.
[18]
H. Ling and K. Okada. 2007. An Efficient Earth Mover's Distance Algorithm for Robust Histogram Comparison. IEEE Trans. Pattern Anal. Mach. Intell. 29, 5, 840--853.
[19]
M-E Nilsback and A. Zisserman. 2006. A Visual Vocabulary for Flower Classification. In IEEE Conf. on Computer Vision and Pattern Recognition, Vol. 2. 1447--1454.
[20]
H. Noltemeier, K. Verbarg, and C. Zirkelbach. 1992. Monotonous bisector* trees - a tool for efficient partitioning of complex scenes of geometric objects. In Data Structures and Efficient Algorithms (vol. 594 of Springer-Verlag Lecture Notes in Computer Science). Berlin, West Germany.
[21]
H. Noltemeier, K. Verbarg, and C. Zirkelbach. 1993. A data structure for representing and efficient querying large scenes of geometric objects: mb*-trees. In Geometric Modelling, Computing/Supplement 8. Springer-Verlag, Vienna, Austria, 211--226.
[22]
O. Pele and M. Werman. 2009. Fast and robust Earth Mover's Distances. In IEEE 12th Intl Conf. on Computer Vision, ICCV, Kyoto, Japan. IEEE Computer Society, 460--467.
[23]
I. R. V. Pola, A. J. M. Traina, and C. Traina. 2009. Easing the Dimensionality Curse by Stretching Metric Spaces. In Scientific and Statistical Database Management. Springer Berlin Heidelberg, Berlin, Heidelberg, 417--434.
[24]
Y. Rubner, C. Tomasi, and L. J. Guibas. 2000. The Earth Mover's Distance as a Metric for Image Retrieval. Int. J. Comput. Vis. 40, 2, 99--121.
[25]
H. Samet. 1983. A quadtree medial axis transform. Commun. ACM 26, 9, 680--693.
[26]
H. Samet. 1985. Reconstruction of quadtrees from quadtree medial axis transforms. Computer vision, graphics, and image processing 29, 3, 311--328.
[27]
H. Samet. 1989. Hierarchical spatial data structures. In Symposium on Large Spatial Databases. Springer, 191--212.
[28]
H. Samet. 2006. Foundations of Multidimensional and Metric Data Structures. Morgan Kaufmann Publishers Inc.
[29]
J. Sankaranarayanan and H. Samet. 2010. Query processing using distance oracles for spatial networks. IEEE Transactions on Knowledge and Data Engineering 22, 8, 1158--1175.
[30]
S. Shirdhonkar and D. W. Jacobs. 2008. Approximate earth mover's distance in linear time. In IEEE Computer Society Conf. on Computer Vision and Pattern Recognition (CVPR), Anchorage, Alaska, USA. IEEE Computer Society.
[31]
E. Tanin, A. Harwood, and H. Samet. 2005. A distributed quadtree index for peer-to-peer settings. In 21st Intl Conf. on Data Engineering (ICDE'05). IEEE, 254--255.
[32]
J. K. Uhlmann. 1991. Satisfying general proximity/similarity queries with metric trees. Inform. Process. Lett. 40, 4, 175--179.
[33]
J. Z. Wang, J. Li, and G. Wiederhold. 2001. SIMPLIcity: Semantics-Sensitive Integrated Matching for Picture LIbraries. IEEE Trans. Pattern Anal. Mach. Intell. 23, 9, 947--963.
[34]
L. Wu and T. Bretschneider. 2004. VP-EMD Tree: An Efficient Indexing Strategy for Image Retrieval. In Proceedings of the Intl Conf. on Imaging Science, Systems and Technology, CISST, Las Vegas, Nevada, USA. CSREA Press, 421--426.
[35]
P. N. Yianilos. 1993. Data structures and algorithms for nearest neighbor search in general metric spaces. In Proceedings of the 4th Annual ACM-SIAM Symposium on Discrete Algorithms. Austin, TX, 311--321.
[36]
C. Zhang, Y. Cai, G. Lin, and C. Shen. 2020. DeepEMD: Few-Shot Image Classification With Differentiable Earth Mover's Distance and Structured Classifiers. In Proc. of the IEEE/CVF Conf. on Computer Vision and Pattern Recognition (CVPR).
Index Terms
- Metric Indexing for the Earth Mover's Distance
Recommendations
Data-Dependent LSH for the Earth Mover’s Distance
STOC 2024: Proceedings of the 56th Annual ACM Symposium on Theory of ComputingWe give new data-dependent locality sensitive hashing schemes (LSH) for the Earth Mover’s Distance (EMD), and as a result, improve the best approximation for nearest neighbor search under EMD by a quadratic factor. Here, the metric EMDs(ℝd,ℓp) consists ...
Indexing Earth Mover’s Distance over Network Metrics
The Earth Mover's Distance (EMD) is a well-known distance metric for data represented as probability distributions over a predefined feature space. Supporting EMD-based similarity search has attracted intensive research effort. Despite the plethora of ...
Parameterized earth mover's distance for efficient metric space indexing
SISAP '11: Proceedings of the Fourth International Conference on SImilarity Search and APplicationsThe Earth Mover's Distance is a well-known distance measure employed in various domains, especially for content-based retrieval in multimedia databases. However, the distance evaluation is a considerably expensive task and thus for large multimedia ...
Comments
Information & Contributors
Information
Published In
November 2023
50 pages
ISBN:9798400703522
DOI:10.1145/3615890
Copyright © 2023 Owner/Author(s).
This work is licensed under a Creative Commons Attribution International 4.0 License.
Sponsors
Publisher
Association for Computing Machinery
New York, NY, United States
Publication History
Published: 11 December 2023
Check for updates
Author Tags
Qualifiers
- Research-article
Funding Sources
- NSF
Conference
GeoSearch '23
Sponsor:
GeoSearch '23: 2nd ACM SIGSPATIAL International Workshop on Searching and Mining Large Collections of Geospatial Data
November 13, 2023
Hamburg, Germany
Contributors
Other Metrics
Bibliometrics & Citations
Bibliometrics
Article Metrics
- 0Total Citations
- 161Total Downloads
- Downloads (Last 12 months)146
- Downloads (Last 6 weeks)19
Reflects downloads up to 15 Jan 2025
Other Metrics
Citations
View Options
Login options
Check if you have access through your login credentials or your institution to get full access on this article.
Sign in