3D resize image for BIG image size

[MAIN QUESTION]

I try to resize (reduce) the image size for further image processing because I can't read whole 3D image data (4176*4176*5856) by small memory. Thus I can't use "imresize3". My another strategy is I can read them separately for all Z and reduce image half in X and Y using "tiffreadVolume" and "imresize". However, it doesn't give me image value that iterpolated in Z. Do you have any idea for this?

[SUB QUESTION]

By trying to code my own "imresize3", I got some examples and coded below but it seems it gives a different result. I guess it is antialiasing process. Do you know how to do antialiasing for images? Is it just filtering?

clc; clear all; close all;

% LOAD IMAGE
s = load('mri');
mriVolumeOriginal = squeeze(s.D);
sizeO = size(mriVolumeOriginal);

% SHOW IMAGE
figure; subplot(131);
slice(double(mriVolumeOriginal),sizeO(2)/2,sizeO(1)/2,sizeO(3)/2);
shading interp, colormap gray; axis image;
title('Original');

% RESIZE IMAGE
f = 0.5; % RESIZE RATIO
mriVolumeResized = imresize3(mriVolumeOriginal,f);
sizeR = size(mriVolumeResized);

% SHOW RESIZED IMAGE (by imresize3)
subplot(132);
slice(double(mriVolumeResized),sizeR(2)/2,sizeR(1)/2,sizeR(3)/2);
shading interp, colormap gray; axis image;
title('Resized (imresize3)');

% NEAREST NEIGHBOR INTERPOLATION (NNI)
% DEFINE THE RESAMPLE SIZE
Col = round(sizeO(1)*f);
Row = round(sizeO(2)*f);
Hig = round(sizeO(3)*f);

%FIND THE RATIO OF THE NEW SIZE BY OLD SIZE
rtR = Row/size(mriVolumeOriginal,1);
rtC = Col/size(mriVolumeOriginal,2);
rtH = Hig/size(mriVolumeOriginal,3);

%OBTAIN THE INTERPOLATED POSITIONS
IR = ceil([1:(size(mriVolumeOriginal,1)*rtR)]./(rtR));
IC = ceil([1:(size(mriVolumeOriginal,2)*rtC)]./(rtC));
IH = ceil([1:(size(mriVolumeOriginal,3)*rtH)]./(rtH));

%ROW_WISE INTERPOLATION
B = mriVolumeOriginal(:,IR,:);

%COLUMN-WISE INTERPOLATION
B = B(IC,:,:);

%HEIGHT-WISE INTERPOLATION
B = B(:,:,IH);

% SHOW RESIZED IMAGE (by nearest neighbor interpolation)
subplot(133);
slice(double(B),sizeR(2)/2,sizeR(1)/2,sizeR(3)/2);
shading interp, colormap gray; axis image;
title('Resized (NNI)');

Answer 1

[Matt J]

My another strategy is I can read them separately for all Z and reduce image half in X and Y using "tiffreadVolume" and "imresize". However, it doesn't give me image value that iterpolated in Z. Do you have any idea for this?

Can't you just imresize with respect to Z in a second step, e.g.,

Image0=rand(500,500,200);

Image1 = imresize(Image0,'OutputSize',[100,100]); %Or, do it slice-by-slice

Image2 = imresize3(Image1,[100,100,40]);  %final image

whos Image*

  Name          Size                     Bytes  Class     Attributes

  Image0      500x500x200            400000000  double              
  Image1      100x100x200             16000000  double              
  Image2      100x100x40               3200000  double              

[Youngju Kim]

I hope to do that but my RAM is small and I can't read my data. I got "out of memory" if I try to load data.

[Matt J]

But you already told us you have enough RAM to read in one image at a time and resize it:

My another strategy is I can read them separately for all Z and reduce image half in X and Y using "tiffreadVolume" and "imresize".

Once you've done this, you can compose these images into Image1 above and resize in Z as shown there.

[Youngju Kim]

I can't read whole 3d data. I used tiffreadVolume to read each image slice (i) of whole images (image.tif).

I'm worried about imresize in Z is not equivalent to imresize3.

path = 'C:\image';
file = 'image.tif';

for i = 1:5856
    img = tiffreadVolume(fullfile(path,file),...
    "PixelRegion",{[1 1 inf], [1 1 inf], [i i]});
    img_bin2 = imresize(img,0.5);
end

Error using tiffreadVolume (line 43)
File "C:\image/image.tif" does not exist.

[Matt J]

I can't read whole 3d data. I used tiffreadVolume to read each image slice (i) of whole images (image.tif).

To implement my proposal, you do not have to read in the whole data at its original size:

path = 'C:\image';

file = 'image.tif';

C=cell(1,5856);

for i = 1:5856

img= tiffreadVolume(fullfile(path,file),...

"PixelRegion",{[1 1 inf], [1 1 inf], [i i]});

C{i} = imresize(img,1/8);

end

C=cat(3,C{:}); [m,n,p]=size(C,3);

img_final=imresize3( C , [m,n,p/8] ); clear C

I'm worried about imresize in Z is not equivalent to imresize3.

They seem to be equivalent in the test below, but why does it matter if there are small differences?

Image0=rand(500,500,200);

Image1 = imresize(Image0,'OutputSize',[100,100]);
Image2a = imresize3(Image1,[100,100,40]);  %final image - version 1

Image2b = imresize3(Image0,[100,100,40]);  %final image - version 2

difference= max(abs(Image2a-Image2b),  [],'all')

difference = 0