Shrink image without using imresize

First, I will say I do not have the Image Processing Toolbox and cannot get it so please do not turn this into yet another debate about why I should use built in functions. I would love to, but I can't for various reasons.

I'm using Matlab 2017a (9.2.0.556344)

I have a video frame that's 1200x1100 (I know, odd dimensions) color RGB image.

I want to end up with a 650x650 color RGB image.

Here's what I tried, which just results in a cropped image. I want a shrunken image. How can I shrink this large image without using imresize or any other function in the Image Processing Toolbox?

v_obj = VideoReader('vid.mp4');

img = readFrame(v_obj);

[Xq,Yq,Zq] = meshgrid(1:1:650,1:1:650,1:1:3);

Vq = interp3(double(img),Xq,Yq,Zq);

Answer 1

[KSSV]

Let I be your image of size 1200x1100.

[m,n,p] = size(I) ; 
[X,Y] = meshgrid(1:m,1:n) ; 
xi = linspace(1,m,650) ; 
yi = linsace(1,n,650) ;
[Xi,Yi] = meshgrid(xi,yi) ; 
Ii = zeros(650,650,3) ; 
for i = 1:p
    Ii(:,:,i) = interp2(X,Y,I(:,:,i),Xi,Yi) ; 
end

[Sean Ford]

Sorry, I should have been clearer

img is a 1200x1100x3 uint8

Here's what I tried based on your response:

[m,n,q] = size(I) ; 
[X,Y,Z] = meshgrid(1:m,1:n,1:q) ; 
xi = linspace(1,m,650); 
yi = linspace(1,n,650); 
zi = linspace(1,q,3); 
[Xi,Yi,Zi] = meshgrid(xi,yi,zi) ; 
Ii = interp3(X,Y,Z,double(I),Xi,Yi,Zi) ; 

I get an error

Error using griddedInterpolant

The grid vectors do not define a grid of points that match the given values.

[KSSV]

Edited the code....

[Sean Ford]

I tried with your edited code KSSV and I get the same error message.

[Image Analyst]

Why does it need to be that exact size? For example you're feeding it into a deep learning network that requires that size?

[KSSV]

Check the dimensions of X, Y, I do they have same rows and columns...?

[Sean Ford]

Image Analyst - It doesn't need to be any exact size, I was just giving an example. This truly is simply shrinking the image to any size I desire. Nothing beyond that. I have a large image, I need to resize to a smaller image (not crop).

KSSV - X and Y have the same number of rows and columns. (1200x1100) each.

[Image Analyst]

Sean:

I used his code and got no errors. Try this:

% Setup demo image.
rgbImage = imread('Peppers.png');
% Resize to 1200 x 1100 - the poster's size.
rgbImage = imresize(rgbImage, [1200, 1100]);

% Display original image.
subplot(1, 2, 1);
imshow(rgbImage);
axis('on', 'image');
title('Original Image', 'FontSize', 20);

% Now do the resizing using interp3().
[rows, columns, numberOfColorChannels] = size(rgbImage)
[X,Y,Z] = meshgrid(1:columns,1:rows,1:numberOfColorChannels);
xi = linspace(1,columns,650); 
yi = linspace(1,rows,650); 
zi = linspace(1,numberOfColorChannels,3); 
[Xi,Yi,Zi] = meshgrid(xi,yi,zi) ; 
resizedImage = uint8(interp3(X,Y,Z,double(rgbImage),Xi,Yi,Zi));
% Display result.
subplot(1, 2, 2);
imshow(resizedImage);
axis('on', 'image');
title('Resized Image', 'FontSize', 20);

[Sean Ford]

That did it! Thanks!!

[Stefan Grandl]

Perfect as always, Image Analyst! Thanks a lot!

Answer 2

[DGM]

Imresize() is currently available in the base toolbox, so you don't need IPT anymore. This change occurred somewhere after R2015b and before (or at) R2019b. I can find no mention of the change in the PDF documentation for either base MATLAB or IPT.

Otherwise, if you need to resize images and you don't have IPT, you can use MIMT imresizeFB(). That is an approximate replacement for imresize(), and supports a similar set of options. One difference of note is that imresizeFB() does not support indexed-color images. It also doesn't support Lanczos kernels.

inpict = imread('peppers.png'); % the source image
imsize(inpict)

ans =

384 512 3 1

outpict = imresizeFB(inpict,[200 400],'bilinear'); % explicit output geometry
imsize(outpict)

ans =

200 400 3 1

outpict = imresizeFB(inpict,[200 NaN],'bilinear'); % implicit output geometry
imsize(outpict)

ans =

200 267 3 1

outpict = imresizeFB(inpict,1.5,'bilinear'); % scaling factor
imsize(outpict)

ans =

576 768 3 1

No loops, no meshgrid, no class handling.

Can you do it directly using interp2()? Yes. If you expect results similar to imresize(), it's more complicated than that -- especially in the case where you're shrinking the image. If the scaling factor is less than 1, imresize() and imresizeFB() perform antialiasing for cleaner output. Depends if that matters to you.