Dr. Timor Kadir (original) (raw)

Affine Invariant Scale Saliency

Overview

The selection of a set of image regions forms the first step in many computer vision algorithms, e.g object recognition or computing image correspondences. The optimal choice for region selection depends ultimately on the application, however, there are three broad classes of image change under which good performance may be required:

1. Global transformations. Features should be repeatable across the expected class of global image transformations. These include both geometric and photometric transformations that arise due to changes in the imaging conditions. For example, we often require that feature segmentations commute with viewpoint change.	Detected regions, denoted by the green ellipses, sould commute with viewpoint change.

We make two contributions. First, we extend the Kadir and Brady region detector to include covariance to affine transformations (Requirement 1), and improve the implementation in a number of ways. Second, we specify a performance measure for Requirements 2 and 3 above, and present comparative results for the Kadir/Brady method, a multi-scale Harris detector and a DoG blob detector.

Code

You can download matlab6 binaries for the full search (ie. slow) affine invariant Scale Saliency algorithm here AffineScaleSaliency_Public_linux_V1.0.tgz, here AffineScaleSaliency_Public_windows_V1.0.tgz, and here AffineScaleSaliency_Public_solaris_V1.0.tgz.

You can download the ground-truth for the intra-class experiments here.

References

Kadir, T. , Zisserman, A. and Brady, M.
An affine invariant salient region detector
Proceedings of the 8th European Conference on Computer Vision, Prague, Czech Republic (2004)
Document: ps.gz PDF

Kadir, T. and Brady, M.
Scale, Saliency and Image description
International Journal of Computer Vision. 45 (2):83-105, November 2001.
Document: ps.gz PDF