A metric for distributions with applications to image databases

@article{Rubner1998AMF,
  title={A metric for distributions with applications to image databases},
  author={Yossi Rubner and Carlo Tomasi and Leonidas J. Guibas},
  journal={Sixth International Conference on Computer Vision (IEEE Cat. No.98CH36271)},
  year={1998},
  pages={59-66}
}
  • Y. Rubner, Carlo Tomasi, L. Guibas
  • Published 4 January 1998
  • Computer Science, Mathematics
  • Sixth International Conference on Computer Vision (IEEE Cat. No.98CH36271)
We introduce a new distance between two distributions that we call the Earth Mover's Distance (EMD), which reflects the minimal amount of work that must be performed to transform one distribution into the other by moving "distribution mass" around. This is a special case of the transportation problem from linear optimization, for which efficient algorithms are available. The EMD also allows for partial matching. When used to compare distributions that have the same overall mass, the EMD is a… 

Figures from this paper

Perceptual metrics for image database navigation
TLDR
A novel approach to the problem of navigating through a collection of images for the purpose of image retrieval is presented, which leads to a new paradigm for image database search and a metric between any two such distributions is defined.
The Earth Mover''s Distance: Lower Bounds and Invariance under Translation
TLDR
This report presents some fast to compute lower bounds on the EMD which may allow a system to avoid exact, more expensive EMD computations during query processing.
Finding color and shape patterns in images
TLDR
This thesis gives a couple of modifications which make the EMD more amenable to partial matching, including the partial EMD in which only a given fraction of the weight in one distribution is forced to match weight in the other, and presents algorithms that are guaranteed to find a globally optimal transformation when matching equal-weight distributions under translation.
Adaptive Color-Image Embeddings for Database Navigation
We present a novel approach to the problem of navigating through a database of color images for the purpose of image retrieval. We endow the database with a metric for the color distributions of the
Using the KL-center for efficient and accurate retrieval of distributions arising from texture images
  • Eric Spellman, B. Vemuri, M. Rao
  • Computer Science
    2005 IEEE Computer Society Conference on Computer Vision and Pattern Recognition (CVPR'05)
  • 2005
TLDR
A novel approach to search such collections of distributions efficiently by using an "average" to represent a large set of objects, thus allowing the search to disregard an unpromising subset with only one comparison to its average.
The Earth Mover's Distance under transformation sets
  • Scott D. Cohen, L. Guibas
  • Computer Science, Mathematics
    Proceedings of the Seventh IEEE International Conference on Computer Vision
  • 1999
TLDR
This work considers the problem of computing a transformation of one distribution which minimizes its EMD to another, and presents a monotonically convergent iteration which can be applied to a large class of EMD under transformation problems, although the iteration may converge to only a locally optimal transformation.
Intensity-based image registration using Earth Mover's Distance
TLDR
Two image alignment measures using Earth Mover's Distance as a metric on the space of joint intensity distributions are introduced and it is shown that EMD-based measures can be efficiently applied to rigid registration tasks.
Earth Mover's Distance based Similarity Search at Scale
TLDR
This paper focuses on optimizing the refinement phase of EMD-based similarity search by adapting an efficient min-cost flow algorithm (SIA) for EMD computation, proposing a dynamic distance bound, and proposed a dynamic refinement order for the candidates which, paired with a concurrent EMD refinement strategy, reduces the amount of needless computations.
Color matching by using tuple matching
  • D. Balthasar
  • Computer Science
    12th International Conference on Image Analysis and Processing, 2003.Proceedings.
  • 2003
TLDR
A new matching method called tuple matching (TM) is presented, which is an algorithm for matching image signatures that are generated from color histograms by using graph theoretical clustering (GT-clustering), which is better than EMD.
Color-Spatial Image Indexing and Applications
TLDR
The color correlogram is proposed as a generic color-spatial indexing tool to tackle various problems that arise in content-based image retrieval and video browsing and a method for hierarchical classification of images via supervised learning is proposed.
...
1
2
3
4
5
...

References

SHOWING 1-10 OF 24 REFERENCES
The Earth Mover's Distance, Multi-Dimensional Scaling, and Color-Based Image Retrieval
TLDR
A novel approach to the problem of navigating through a database of color images by using multi-dimensional scaling techniques to embed a group of images as points in a two or three-dimensional Euclidean space so that their distances are preserved as much as possible.
QBIC project: querying images by content, using color, texture, and shape
TLDR
The main algorithms for color texture, shape and sketch query that are presented, show example query results, and discuss future directions are presented.
Similarity of color images
TLDR
Two new color indexing techniques are described, one of which is a more robust version of the commonly used color histogram indexing and the other which is an example of a new approach tocolor indexing that contains only their dominant features.
A distance metric for multidimensional histograms
The analysis of proximities: Multidimensional scaling with an unknown distance function. I.
TLDR
The program is proposed as a tool for reductively analyzing several types of psychological data, particularly measures of interstimulus similarity or confusability, by making explicit the multidimensional structure underlying such data.
A multi-channel filtering approach to texture segmentation
  • F. Farrokhnia, Anil K. Jain
  • Computer Science
    Proceedings. 1991 IEEE Computer Society Conference on Computer Vision and Pattern Recognition
  • 1991
TLDR
An integrated segmentation technique that combines the strengths of the previous two techniques while eliminating their weaknesses is proposed and is truly unsupervised, since it eliminates the need for knowing the exact number of texture categories in the image.
Texture Features for Browsing and Retrieval of Image Data
TLDR
Comparisons with other multiresolution texture features using the Brodatz texture database indicate that the Gabor features provide the best pattern retrieval accuracy.
Image Representation Using 2D Gabor Wavelets
  • T. Lee
  • Computer Science
    IEEE Trans. Pattern Anal. Mach. Intell.
  • 1996
TLDR
The conditions under which a set of continuous 2D Gabor wavelets will provide a complete representation of any image are derived, and self-similar wavelet parametrization is found which allow stable reconstruction by summation as though the wavelets formed an orthonormal basis.
Generalized texture representation and metric
Multidimensional scaling by optimizing goodness of fit to a nonmetric hypothesis
TLDR
The fundamental hypothesis is that dissimilarities and distances are monotonically related, and a quantitative, intuitively satisfying measure of goodness of fit is defined to this hypothesis.
...
1
2
3
...