Applies a linear or nonlinear transformation to an image I.
Takes the center of the image as the origin, not the top left corner.
Also, the coordinate system is row/column format, so H must be also.
The bounding box of the image is set by the BBOX argument, a string that
can be 'loose' (default) or 'crop'. When BBOX is 'loose', J includes the
whole transformed image, which generally is larger than I. When BBOX is
'crop' J is cropped to include only the central portion of the
transformed image and is the same size as I. The 'loose' flag is
currently inexact (because of some padding/cropping). Preserves I's type.
USAGE
J = imtransform2( I, H, varargin )
INPUTS - common
I - input image [converted to double]
H - 3x3 nonsingular homography matrix
varargin - additional params (struct or name/value pairs)
.method - ['linear'] 'nearest', 'spline', 'cubic' (for interp2)
.bbox - ['crop'] or 'loose'
.show - [0] figure to use for optional display
.pad - [0] padding value (scalar, 'replicate' or 'none')
.useCache - [0] optionally cache precomp. for given transform/dims.
.us - [] can specify source r/c for each target (instead of H)
.vs - [] can specify source r/c for each target (instead of H)
OUTPUTS
J - transformed image
EXAMPLE - rigid transformation (rotation + translation)
I=imread('peppers.png');
R = rotationMatrix(pi/4); T=[1; 3]; H=[R T; 0 0 1];
J = imtransform2(I,H,'show',1,'pad','replicate');
EXAMPLE - general homography (out of plane rotation)
load trees; I=X;
S=eye(3); S([1 5])=1/500; % zoom out 500 pixels
H=S^-1*rotationMatrix([0 1 0],pi/4)*S;
J = imtransform2(I,H,'bbox','loose','show',1);
EXAMPLE - rotation using three approaches (and timing)
load trees; I=imResample(X,4); angle=35; method='bilinear';
% (1) rotate using imrotate (slow)
tic; J1 = imrotate(I,angle,method,'crop'); toc
% (2) rotate using a homography matrix
R=rotationMatrix(angle/180*pi); H=[R [0; 0]; 0 0 1];
tic; J2 = imtransform2(I,H,'bbox','crop','method',method); toc
% (3) rotate by explicitly specifying target rs/cs
m=size(I,1)+4; n=size(I,2)+4; m2=(m-1)/2; n2=(n-1)/2;
[cs,rs]=meshgrid(-n2:n2,-m2:m2); vs=R*[cs(:) rs(:)]';
us=reshape(vs(2,:),m,n)-rs; vs=reshape(vs(1,:),m,n)-cs;
tic, J3=imtransform2(I,[],'us',us,'vs',vs,'method',method); toc
% compare all results
figure(1); clf; subplot(3,2,1); im(I); subplot(3,2,2); im(J1);
subplot(3,2,3); im(J2); subplot(3,2,4); im(abs(J1-J2)); title('J1-J2')
subplot(3,2,5); im(J3); subplot(3,2,6); im(abs(J2-J3)); title('J2-J3')
See also TEXTUREMAP, INTERP2
Piotr's Computer Vision Matlab Toolbox Version 3.01
Copyright 2014 Piotr Dollar. [pdollar-at-gmail.com]
Licensed under the Simplified BSD License [see external/bsd.txt]