Image and Video Analysis

Hough Pyramid Matching for Large Scale Image Retrieval

Inspired by pyramid matching and Hough transform we present a relaxed and flexible spatial matching model which applies the concept of pyramid match to the transformation space. It is invariant to similarity transformations and free of inlier-count verification. Being linear in the number of correspondences, it is fast enough to be used for re-ranking in large scale image retrieval. It imposes one-to-one mapping and is flexible, allowing non-rigid motion and multiple matching surfaces or objects.

Given sets of local features and visual words assigned using a codebook, Hough Pyramid Matching (HPM) is applied on the tentative correspondences that are formed. A strength is finally assigned to each correspondence reflecting weather it is geometrically consistent with other correspondences. Final pairwise image similarity is the sum of all strengths.

The code has been developed in C++. Input files are local features and corresponding visual words for database and query images. The provided output is the ranked list of images for all queries to be used for comparisons with the proposed approach.

The binary code is available for Linux 64-bit platform. No installation is needed. The documentation provides a guide to appropriate commands and options.

Licence / Citation

The software may be used freely for research purposes only. Please cite our ICCV paper below if you use it.


Version 0.3 (05/03/2012), Linux 64-bit binary and documentation (tar.gz, 140 KB)


Contact Giorgos Tolias for any details.



G. Tolias, Y. Avrithis. Speeded-up, Relaxed Spatial Matching. In Proceedings of International Conference on Computer Vision (ICCV 2011), Barcelona, Spain, November 2011.
[ Abstract ]
[ Bibtex ] [ PDF ]