How to estimate homography using RANSAC on SIFT mathing results

Question

In image matching, with Matlab, I found a vector of correspondences of two images using Sift and now I have to estimate the homography matrix. Any simple way to do it? Thanks in advance

Answer 1

The book "Multiple View Geometry in Computer Vision" by Richard Hartley and Andrew Zisserman is an excellent resource for this sort of problem, and helpfully they also provide a set of Matlab functions to perform common tasks. Look at this page and download the file vgg_H_from_x_lin.m, which estimates the homography between two sets of points using a linear method. Here's an example (with some completely made-up numbers):

x1 = [ 10 20; 13 23; 45 35 ].';
x2 = [ 103 301; 106 305; 80 229 ].';
H = vgg_H_from_x_lin(x1, x2);

Alternatively, download the file ransacfithomography_vgg.m to compute the homography using RANSAC. This time you must give an inlier threshold for RANSAC.

x1 = [ 10 20; 13 23; 45 35 ].';
x2 = [ 103 301; 106 305; 80 229 ].';
H = ransacfithomography_vgg(x1, x2, 0.02);

Answer 2

The simplest way to do this is to use the estimateGeometricTransform function in the Computer Vision System Toolbox.