Thread Subject: imtransform question

Dear all,

probably an easy question for the diehards but I'm struggling a while already.

I'm trying to transform an image using imtransform. The main idea is that I've got a large image, which I will register/match to another image. Therefore I just select a small ROI to perform the registration (using an optimization). Subsequently, this found transformation is applied to the entire (large) image. However, the resulting image is not matched, even if rotations were not taken into account.

To demonstrate my problem, I simplified my code to the following:
I. first cropping, then imtransform,
II. first imtransform (use the same transformation), then cropping.

Why doesn't give the two (apparently similar) routes the same result? And how can I solve this?

Many thanks in advance,
Kees.

%***************************************************************
function main()
Img = zeros(400,500); Img(100:200,200:400) = 1; % test image

ClipBox = [50,size(Img,1)-50;50,size(Img,2)-50]; %[vert0,vert1,hor0.hor1]
dx = 100;
dy = -100;
dr = 0;
Ax1 = 1:size(Img,1); % coordinates of the axes
Ax2 = 1:size(Img,2);

% Transform image and then do clipbox
ImgT0 = TransformImage(Img, dx,dy,dr, Ax1,Ax2);
ImgT0 = ImgT0(ClipBox(1,1):ClipBox(1,2),ClipBox(2,1):ClipBox(2,2));

% Do clipbox and then transform image
Img = Img(ClipBox(1,1):ClipBox(1,2),ClipBox(2,1):ClipBox(2,2));
Ax1 = Ax1(ClipBox(1,1):ClipBox(1,2));
Ax2 = Ax2(ClipBox(2,1):ClipBox(2,2));
ImgT1 = TransformImage(Img, dx,dy,dr, Ax1,Ax2);

figure(2);
subplot(3,2,1); imagesc(Img); colormap(gray); title('original image cropped')
subplot(3,2,3); imagesc(ImgT0); colormap(gray); title('transformed image then cropped')
subplot(3,2,5); imagesc(ImgT1); colormap(gray); title('cropped original image then transformed')
subplot(3,2,6); imagesc(ImgT0-ImgT1); colormap(gray); title('difference two transf. images');


function ImgT = TransformImage(Img, dx,dy,dr, Ax1, Ax2)
udata = [Ax1(1), Ax1(end)];
vdata = [Ax2(1), Ax2(end)];

tform = maketform('affine',[cos(dr/180*pi) -sin(dr/180*pi) 0; ...
                            sin(dr/180*pi) cos(dr/180*pi) 0; ...
                            dx dy 1]);
[ImgT] = imtransform(Img, tform, 'linear', ...
                     'udata', udata,'vdata', vdata,...
                     'xdata', udata,'ydata', vdata,...
                     'size', size(Img), 'fill', NaN);

%***************************************************************

On 14 Okt., 14:18, "kees de Kapper" <kees_de_kap...@hotmail.com>
wrote:
> Dear all,
>
> probably an easy question for the diehards but I'm struggling a while already.
>
> I'm trying to transform an image using imtransform. The main idea is that I've got a large image, which I will register/match to another image. Therefore I just select a small ROI to perform the registration (using an optimization). Subsequently, this found transformation is applied to the entire (large) image. However, the resulting image is not matched, even if rotations were not taken into account.
>
> To demonstrate my problem, I simplified my code to the following:
> I. first cropping, then imtransform,
> II. first imtransform (use the same transformation), then cropping.
>
> Why doesn't give the two (apparently similar) routes the same result? And how can I solve this?
>
> Many thanks in advance,
> Kees.
>
> %***************************************************************
> function main()
> Img = zeros(400,500); Img(100:200,200:400) = 1; % test image
>
> ClipBox = [50,size(Img,1)-50;50,size(Img,2)-50]; %[vert0,vert1,hor0.hor1]
> dx = 100;
> dy = -100;
> dr = 0;
> Ax1 = 1:size(Img,1); % coordinates of the axes
> Ax2 = 1:size(Img,2);
>
> % Transform image and then do clipbox
> ImgT0 = TransformImage(Img, dx,dy,dr, Ax1,Ax2);
> ImgT0 = ImgT0(ClipBox(1,1):ClipBox(1,2),ClipBox(2,1):ClipBox(2,2));
>
> % Do clipbox and then transform image
> Img = Img(ClipBox(1,1):ClipBox(1,2),ClipBox(2,1):ClipBox(2,2));
> Ax1 = Ax1(ClipBox(1,1):ClipBox(1,2));
> Ax2 = Ax2(ClipBox(2,1):ClipBox(2,2));
> ImgT1 = TransformImage(Img, dx,dy,dr, Ax1,Ax2);
>
> figure(2);
> subplot(3,2,1); imagesc(Img); colormap(gray); title('original image cropped')
> subplot(3,2,3); imagesc(ImgT0); colormap(gray); title('transformed image then cropped')
> subplot(3,2,5); imagesc(ImgT1); colormap(gray); title('cropped original image then transformed')
> subplot(3,2,6); imagesc(ImgT0-ImgT1); colormap(gray); title('difference two transf. images');
>
> function ImgT = TransformImage(Img, dx,dy,dr, Ax1, Ax2)
> udata = [Ax1(1), Ax1(end)];
> vdata = [Ax2(1), Ax2(end)];
>
> tform = maketform('affine',[cos(dr/180*pi) -sin(dr/180*pi) 0; ...
>                             sin(dr/180*pi) cos(dr/180*pi) 0; ...
>                             dx dy 1]);
> [ImgT] = imtransform(Img, tform, 'linear', ...
>                      'udata', udata,'vdata', vdata,...
>                      'xdata', udata,'ydata', vdata,...
>                      'size', size(Img), 'fill', NaN);
>
> %***************************************************************

Have you tried the cpselect tool? It's really good for finding
Transformation points.

thanks for replying. CP select has a different purpose then greyscale pixel registration that I want to do. I don't want to (can't) detect clear features. Moreover it won't be to accurate enough.


HB <mister.hb@gmail.com> wrote in message <01ecfe7b-aedd-4428-a4b4-904d2f06d80c@i19g2000yqa.googlegroups.com>...
> Have you tried the cpselect tool? It's really good for finding
> Transformation points.

kick

"kees de Kapper" <kees_de_kapper@hotmail.com> wrote in message <j799al$qd6$1@newscl01ah.mathworks.com>...
>
> Why doesn't give the two (apparently similar) routes the same result? And how can I solve this?
==========

Because you have the roles of Ax1 and Ax2 reversed. Ax1 corresponds to the y axis while Ax2 corresponds to the x axis. Simply change the signature of Transform Image to

function ImgT = TransformImage(Img, dx,dy,dr, Ax2, Ax1)

and you will see that the problem is fixed.

Kick

Matt J, thanks! that was the trick!
Although I'm reasonably experienced with Matlab, I have been struggling with this issue for a while. Now, it is finally solved. So easy!

/Kees