Wednesday, April 22, 15:00 - 17:30, Location: Show and Tell Area B
Guoshen Yu, Jean-Michel Morel
The aim of the demo is to illustrate with direct image and video comparison experiments a new method to compare images with a fully affine invariant algorithm. Pairs of images and videos selected by the participants in a large database will be compared automatically to check whether they contain views of identical or similar objects. Striking examples of success, but also some instructive failure cases will be commented upon.
Why is this important?
If a physical object has a smooth or piecewise smooth boundary, its images obtained by cameras in varying positions undergo smooth apparent deformations. These deformations are locally well approximated by affine transforms of the image plane. These affine deformations from one image to another can be very strong. Each deformation can be quantified by its transition tilt, a new concept introduced by the authors. The experiments will illustrate what a large transition tilt means in practice.
The method shown in the demo, Affine-SIFT (ASIFT), simulates all image views obtainable by varying the two camera axis orientation parameters, namely the latitude and the longitude angles, left over by the classic SIFT method. Then it covers the other four parameters by using the SIFT method itself.
Against any prognosis, simulating all views depending on the two camera orientation parameters is feasible with no dramatic computational load: A two-resolution scheme further reduces the ASIFT complexity to about twice that of SIFT. Comparisons with SIFT and other affine comparisons methods will also been shown on demand.