How can I rescale an image while not changing the dimension of the image?

How can I rescale an image while not changing the dimension of the image? When I scale it up, it just crops the image to fit within the original size. When I scale it down, it pads the edges of the image with zeros. I want to zoom in and out about the center of the image. When using "imresize", it changes the dimensions of the image to preserve all the content. I have the image processing toolbox.

[karim botros]

I guess you have two image with similar background one of them the object/scene looks smaller than the other,

The problem is more of a logical computer vision problem not a matlab one. but matlab is always there to helps us.

solution:

Crop the scene/object in both images to be able to get similar sizes before resizeing, then resize one of the images to the size of the other image.

important: this code is not optimized, the main aim is to explain the idea, it can be written in a more efficient way and less lines, nevertheless it won't be clear if i did so.

try to run cropping of image one at a time, so you can handle the cropping,

to confirm the cropping right click then crop image or double click.

Matlab code:

im0 = imread('yourimage0.png');
im1 = imread('yourimage1.png');

%%
figure;
imshow(im0)
cropped0 = imcrop(im0);
%%
figure(1);
imshow(im1)
cropped1 = imcrop(im1);

im0f = imresize(cropped0,[size(cropped1,1) size(cropped1,2)]);
im1f = cropped1;


imwrite(im0f,'image0r.png')
imwrite(im1f,'image1r.png')

Answer 1

[Image Analyst]

I'm not sure what you're talking about. How are you performing the scaling? Are you using the zoom() function? Are you using 'InitialMagnification' property of imshow()?

[Sonoma Rich]

For example, I have a 640x512 image. I want to rescale it up or down, but I still want the processed image to be 640x512. If I scale it up, it crops the rescaled image to fit within 640x512. If I scale it down, it pads the image with zeros to fit within 640x512. Similar to a digital zoom on a video camera.

[Image Analyst]

Assuming you're doing your scaling with zoom(), this does not affect what you process. It only affects what is displayed. The display region on your monitor is the same number of pixels of course but calling zoom() will replicate pixels for display, but only for display. You can click and drag the image to move it around once it's zoomed. If you zoom way out, so that your image gets tiny on the screen, I don't know what it does - if it just shrinks the axes control, or if it keeps the axes the same size and just makes it black outside the tiny image. But it doesn't really matter - your image is still the same variable and all this is just for display only. It won't affect what parts of the image that you process.

[Sonoma Rich]

This is not exact what I need. The zoom function assumes I am displaying the image. I need to do further processing of the scaled image. In addition, it can be larger than the monitor as well as have different bit depths. I am looking for something similar to imresize which outputs a processed image.

[Walter Roberson]

imresize() does output the processed image ?

[Image Analyst]

Like Walter says, imresize() does output a processed (i.e. resized) image. You can then cast that to a different bit depth, if you want. Look:

smallImage = imread('pout.tif');
whos smallImage
% Enlarge the image to 4000 by 6000, bigger than the monitor.
bigImage = uint16(imresize(smallImage, [4000,6000]));
whos bigImage

    Name              Size             Bytes  Class    Attributes
    smallImage      291x240            69840  uint8 
    Name             Size                 Bytes  Class     Attributes
    bigImage      4000x6000            48000000  uint16

No zooming or displaying was done at all. Please tell us why this does not meet your needs.

[Image Analyst]

Richard - are you still there?????

[Image Analyst]

I don't understand. Simply use imcrop(). What's the difficulty?

Answer 2

[Jurgen]

I think you want to look into imtransform, specifically the Xdata and Ydata properties. E.g. you can rotate around the center--keeping original size. Or rotate around the left corner (image moves out of view if you keep original dataspace). Etcetera for other 'focus' points and transform types.

[Image Analyst]

imrotate() can also crop or enlarge the canvass when you rotate. But he didn't say anything about rotation, just scaling. I'm still trying to figure out what he wants though.

[Jurgen]

e.g. means for example, or exempli gratia according to wiki--since we all know what is meant by rotate. Based on his wording, richard may not differentiate properly between scaling and cropping, so I chose rotate as example; imtransform can resize and crop too.

I think what he means is what some photo-viewing applications do when you scroll: the display window size remains constant and the image is scaled and cropped(or padded) around the center to the windowsize.

Should also be possible with imresize, but after the resize the user would need to manually crop or padd, e.g. using padarray or imcrop.

Answer 3

[dzsuz87]

I think this guy wanted something very similar to my actual problem.

Now I needed a function which can zoom in/out of an image, and keeps the original size of the matrix. The second requirement was that the center pixel should remain at the same position. This is what I use now:

function OutPicture = HardZoom(InPicture, ZoomFactor)

      if ZoomFactor == 0
          OutPicture = [];
          return;
      end
      if ZoomFactor < 0
          InPicture = rot90(InPicture, 2);
          ZoomFactor = 0 - ZoomFactor;
      end
      if ZoomFactor == 1
          OutPicture = InPicture;
          return;
      end
      xSize = size(InPicture, 1);
      ySize = size(InPicture, 2);
      xCrop = xSize / 2 * abs(ZoomFactor - 1);
      yCrop = ySize / 2 * abs(ZoomFactor - 1);
      zoomPicture = imresize(InPicture, ZoomFactor);
      if ZoomFactor > 1
          OutPicture = zoomPicture( 1+xCrop:end-xCrop, 1+yCrop:end-yCrop );
      else
          OutPicture = zeros(xSize, ySize);
          OutPicture( 1+xCrop:end-xCrop, 1+yCrop:end-yCrop ) = zoomPicture;
      end

end

Answer 4

[Victor Leborán Alvarez]

Corrected version that avoids the problem of the even/odd dimensions, with the previous solution the code fails when the dimensions of x or y are even, due to the division by 2. It also works with RGB images of MxNx3. Also I've renamed x and y, because in matlab the first dimension is the height of the image so I've used

ySize = vertical height.

xSize= horiz width.

% Función que permite hacer zoom del contenido pero sin variar las dimensiones de la matriz
function OutPicture = HardZoom(InPicture, ZoomFactor)
      % Si el factor de escala es 1, no se hace nada
      if ZoomFactor == 1
          OutPicture = InPicture;
          return;
      end
      % Se obtienen las dimensiones de las imágenes
      ySize = size(InPicture, 1);
      xSize = size(InPicture, 2);
      zSize = size(InPicture, 3);
      yCrop = floor(ySize / 2 * abs(ZoomFactor - 1));
      xCrop = floor(xSize / 2 * abs(ZoomFactor - 1));
      % Si el factor de escala es 0 se devuelve una imagen en negro
      if ZoomFactor == 0
          OutPicture = uint8(zeros(ySize, xSize, zSize));
          return;
      end
      % Se reescala y se reposiciona en en centro
      zoomPicture = imresize(InPicture, ZoomFactor);
      ySizeZ = size(zoomPicture, 1);
      xSizeZ = size(zoomPicture, 2);      
      if ZoomFactor > 1
          OutPicture = zoomPicture( 1+yCrop:yCrop+ySize, 1+xCrop:xCrop+xSize, :);
      else
          OutPicture = uint8(zeros(ySize, xSize, zSize));
          OutPicture( 1+yCrop:yCrop+ySizeZ, 1+xCrop:xCrop+xSizeZ, :) = zoomPicture;
      end
end