Homography Matrix

Can somebody please help me in understanding how to calculate an homography matrix in matlab.

1 Comment

Lalit Patil
Lalit Patil on 22 Jan 2013
I am finding homography matrix in another way and it is 3*3
So, i want to know that in image where to apply or what to do of this homography matrix.?
Why we are finding this matrix..?

Sign in to comment.

Answers (2)

David Young
David Young on 15 Jan 2012
You don't say what you are starting from. If you have a set of matched input and output points, one possible method is given here. A simple implementation is below.
function v = homography_solve(pin, pout)
% HOMOGRAPHY_SOLVE finds a homography from point pairs
% V = HOMOGRAPHY_SOLVE(PIN, POUT) takes a 2xN matrix of input vectors and
% a 2xN matrix of output vectors, and returns the homogeneous
% transformation matrix that maps the inputs to the outputs, to some
% approximation if there is noise.
%
% This uses the SVD method of
% http://www.robots.ox.ac.uk/%7Evgg/presentations/bmvc97/criminispaper/node3.html
% David Young, University of Sussex, February 2008
if ~isequal(size(pin), size(pout))
error('Points matrices different sizes');
end
if size(pin, 1) ~= 2
error('Points matrices must have two rows');
end
n = size(pin, 2);
if n < 4
error('Need at least 4 matching points');
end
% Solve equations using SVD
x = pout(1, :); y = pout(2,:); X = pin(1,:); Y = pin(2,:);
rows0 = zeros(3, n);
rowsXY = -[X; Y; ones(1,n)];
hx = [rowsXY; rows0; x.*X; x.*Y; x];
hy = [rows0; rowsXY; y.*X; y.*Y; y];
h = [hx hy];
if n == 4
[U, ~, ~] = svd(h);
else
[U, ~, ~] = svd(h, 'econ');
end
v = (reshape(U(:,9), 3, 3)).';
end
If your initial data is in some other form, such as camera position parameters relative to the plane, please say.
EDIT - added:
To apply the resulting matrix to a set of points, you can use the following function.
function y = homography_transform(x, v)
% HOMOGRAPHY_TRANSFORM applies homographic transform to vectors
% Y = HOMOGRAPHY_TRANSFORM(X, V) takes a 2xN matrix, each column of which
% gives the position of a point in a plane. It returns a 2xN matrix whose
% columns are the input vectors transformed according to the homography
% V, represented as a 3x3 homogeneous matrix.
q = v * [x; ones(1, size(x,2))];
p = q(3,:);
y = [q(1,:)./p; q(2,:)./p];
end

13 Comments

Sean Lawson
Sean Lawson on 27 Mar 2012
Hi
I have a question regarding the final transform matrix v.
I want to apply the homography(projective transform) in Matlab, I use this approach to calculate the matrix v. However, v(3,3) ~= 1, whereas Matlab requires v(3,3) = 1, do you know how to fix this? Or just v/v(3,3) works?
David Young
David Young on 30 Mar 2012
Hi Sean, I'm adding a function to my answer with I hope deals with this question.
Royi Avital
Royi Avital on 30 Jul 2012
What if you want to impose Affine Matrix (Last row to be [0, 0, 1])?
Hi, First of all thanks for the code it really helped. But can you explain to me why exactly are u reshaping only the last column. Im new to matlab so I dont really pick up on things fast. v = (reshape(U(:,9), 3, 3)).'; ?? Thanks, Jason
Erez Farhan
Erez Farhan on 14 Jun 2013
Hi, David- thanks for your post. Please note: I believe your solution doesn't account for normalization of the homography matrix- which can be done by dividing it by it's second singular value.
pankhuri
pankhuri on 19 Jun 2013
can this be applied for 2D to 3D projection??
i want to know the difference between transformation matrix and homography matrix
Tianya Zhang
Tianya Zhang on 28 Nov 2017
Edited: Tianya Zhang on 28 Nov 2017
svd() function returns U, S and V. h can be obtained from the last colum of V. Why you use U to get h in your code? That's confusing.... Is that an obvious mistake?
Also curious about Tianya's comment above!
"svd() function returns U, S and V. h can be obtained from the last colum of V. Why you use U to get h in your code? That's confusing.... Is that an obvious mistake?"
Albert Dayn
Albert Dayn on 13 Feb 2019
In response to Tianya and Edwin's question, U is correct. It looks like in this function, h is constructed as the transpose of the usual matrix (each point is represented by 2 columns here, instead of 2 rows). If it were constructed as the link (for the SVD method) specifies, you would indeed use the last column of V.
how to get target physical coordinates in millimetres
Wei Han
Wei Han on 25 Mar 2022
Edited: Wei Han on 25 Mar 2022
Hi, I just want to thank for your code, it's really helped me!
This is very helpful. I think equation for q needs to be slightly altered.
q = v * [x; y; ones(1, size(x,2))];

Sign in to comment.

Matt J
Matt J on 25 Apr 2022
Edited: Matt J on 25 Apr 2022

0 votes

1 Comment

Matt J
Matt J on 26 Apr 2022
Edited: Matt J on 26 Apr 2022
You shouldn't be extracting the matrix. You should be using tform.transformPointsForward().

Sign in to comment.

Categories

Find more on Sparse Matrices in Help Center and File Exchange

Asked:

on 15 Jan 2012

Edited:

on 26 Apr 2022

Community Treasure Hunt

Find the treasures in MATLAB Central and discover how the community can help you!

Start Hunting!