How to correct the skewed perspective of an image

I'm trying to transform this image:

into something like this:

basically by matching the corners of the blue square with the corners of the image and making every other pixel transform accordingly.

I can find the corners of the distorted blue square and its centroid, its border and diagonals, if any of these can be useful for the solution. (Here some intermediate steps I did to find them: http://imgur.com/a/pc7VB )

I thought that I could create a projective tform based on matching corners. My attempt was:

img = imread('img1.png');

if size(img,3)==3
   img = rgb2gray(img);
end

movingPoints = [12 17; 253 16; 269 259;16 256]; %coordinate of distorted corners 
fixedPoints=[0 0;size(img,1) 0;size(img,1) size(img,2);0 size(img,2)]; %coordinate of image's corners 
TFORM = fitgeotrans(movingPoints,fixedPoints,'projective');
R=imref2d(size(img),[1 size(img,2)],[1 size(img,1)]);

imgTransformed=imwarp(imread('img1.png'),TFORM,'OutputView',R);
figure, imshow(imgTransformed);

But it seems to accomplish almost nothing. I also tried with:

TFORM = fitgeotrans(movingPoints,fixedPoints,'affine')

TFORM = fitgeotrans(movingPoints,fixedPoints,'polynomial',2)

But still no result. I know the OpenCV functions getperspectivetransform and warpPerspective can do something like this. Is there a way to do it in MATLAB?

[Nikhil Sreekumar]

Hi Jacopo,

The movingPoints, is it forming a quadrilateral or a rectangle or a curved polygon? On trying to use the coordinates for the movingPoints as [25 17; 252 9; 278 275;-3 239] with affine transformation, I was able to warp the image into the required shape. Am I missing something here?

Thanks

Nikhil

Answer 1

[DGM]

It's always a challenge trying to keep track of which MATLAB/IPT tools use image axes coordinates [x y], and which ones use array coordinates [y x]. in this case, the coordinates were flipped and one corner wasn't very accurately placed.

img = imread('https://www.mathworks.com/matlabcentral/answers/uploaded_files/159518/image.png');
sz = size(img);

movingPoints = [17 12;
                7 254; 
                259 269;
                256 16]; %coordinate of distorted corners [x y]
fixedPoints = [0 0;
               0 sz(1);
               sz(2) sz(1);
               sz(2) 0]; %coordinate of image's corners [x y]

TFORM = fitgeotrans(movingPoints,fixedPoints,'projective');
R = imref2d(sz,[1 sz(2)],[1 sz(1)]);
imgTransformed = imwarp(img,TFORM,'OutputView',R);

imshow(imgTransformed);