Convert image into Patches of size 64*64 and get each patch

Hello, I have RGB image dataset.I want to convert an image into patches and save each patch.How to do this in matlab. The patch size should be 64*64

[Rajesh Das]

I = imread('000000_10.png');
imSz = size(I);
patchSz = [64 64];
xIdxs = [1:patchSz(2):imSz(2) imSz(2)+1];
yIdxs = [1:patchSz(1):imSz(1) imSz(1)+1];
patches = cell(length(yIdxs)-1,length(xIdxs)-1);
for i = 1:length(yIdxs)-1
    Isub = I(yIdxs(i):yIdxs(i+1)-1,:);
    for j = 1:length(xIdxs)-1
        patches{i,j} = Isub(:,xIdxs(j):xIdxs(j+1)-1);
    end
end

###try this###

[Walter Roberson]

@Rajesh Das

Needs adjustment for RGB.

[Image Analyst]

@Rajesh Das, and should be down below in the official "Answer" section with all the other answers of this question.

Answer 1

[Sven]

Hi Tahir, try this:

I = imread('rice.png');
imSz = size(I);
patchSz = [64 64];
xIdxs = [1:patchSz(2):imSz(2) imSz(2)+1];
yIdxs = [1:patchSz(1):imSz(1) imSz(1)+1];
patches = cell(length(yIdxs)-1,length(xIdxs)-1);
for i = 1:length(yIdxs)-1
    Isub = I(yIdxs(i):yIdxs(i+1)-1,:);
    for j = 1:length(xIdxs)-1
        patches{i,j} = Isub(:,xIdxs(j):xIdxs(j+1)-1);
    end
end

This produces a cell of your 64x64 patches:

>> patches
patches = 
  [64x64 uint8]    [64x64 uint8]    [64x64 uint8]    [64x64 uint8]
  [64x64 uint8]    [64x64 uint8]    [64x64 uint8]    [64x64 uint8]
  [64x64 uint8]    [64x64 uint8]    [64x64 uint8]    [64x64 uint8]
  [64x64 uint8]    [64x64 uint8]    [64x64 uint8]    [64x64 uint8]

And you can access each one:

figure, imagesc(patches{2,3})

Did that do what you wanted?

Thanks, Sven.

[Tahir]

Yes, Thank you.

[Ahmad]

hello. its work well on 256 by 256 image but i have 512 by 512 image. it gives the following error. Index exceeds matrix dimensions.

[Rajesh Das]

What will be the code if instead of RGB image its a Grayscale image?

[kashif Mobeen]

Hello Sven,

I've tried this code, I gave RGB image as an input but it is returning patches in a grayscale. How this code will work, if we need patches in RGB as well??

[Walter Roberson]

I = imread('flamingos.jpg');
imSz = size(I);
patchSz = [64 64];
xIdxs = [1:patchSz(2):imSz(2) imSz(2)+1];
yIdxs = [1:patchSz(1):imSz(1) imSz(1)+1];
patches = cell(length(yIdxs)-1,length(xIdxs)-1);
for i = 1:length(yIdxs)-1
    Isub = I(yIdxs(i):yIdxs(i+1)-1,:,:);
    for j = 1:length(xIdxs)-1
        patches{i,j} = Isub(:,xIdxs(j):xIdxs(j+1)-1,:);
    end
end

[Neha Rathore]

Hi! The code worked well. How can i recombine these patches to form an original image.

Answer 2

[Image Analyst]

This is answered by the FAQ in two different ways (mat2cell and indexing). Take your pick as to which is easier, more convenient, and intuitive for you.

http://matlab.wikia.com/wiki/FAQ#How_do_I_split_an_image_into_non-overlapping_blocks.3F

[Ade Aulya]

hi, what if I wanted to get the glcm of each patches automatically ? im so sorry, this is the first time im work on matlab.

I have following this code as mentioned and adding code for extracting GLCM. it works to get the patches but I still make mistake when I ad glcm code. still get error in :

Error in GLCM1 (line 53)
        baseFileName = ca(k).name;

here's the code that I've followed :

folder = fullfile('D:\ProgramFiles\MATLAB\R2017a\bin\DATABC\Train\G1');
  fileNames = dir(fullfile(folder, '*.tif'));
  baseFileName1 = '6a.tif';
  fullFileName = fullfile(folder, baseFileName1); 
  if ~exist(fullFileName, 'file') 
  fullFileName = baseFileName1; 
  if ~exist(fullFileName, 'file') 
  errorMessage = sprintf('Error: %s does not exist.', fullFileName);
  uiwait(warndlg(errorMessage));
  return;
  end
  end

    h=zeros(28,4);%row->no. of image, column->no. of feature 

    % Create blocks
    rgbImage = imread(fullFileName);
    gr = rgb2gray(rgbImage);
    imshow(gr);
    set(gcf, 'units','normalized','outerposition',[0 0 1 1]);
    drawnow;
    [rows, columns, numberOfColorBands] = size(gr);
    blockSizeR = 100; % Rows in block.
    blockSizeC = 100; % Columns in block.
    % The size of each block in rows. 
    wholeBlockRows = floor(rows / blockSizeR);
    blockVectorR = [blockSizeR * ones(1, wholeBlockRows), rem   (rows,blockSizeR)];
    % The size of each block in columns. 
    wholeBlockCols = floor(columns / blockSizeC);
    blockVectorC = [blockSizeC * ones(1, wholeBlockCols), rem(columns,   blockSizeC)];
    % Create the cell array
    % Divided up into blocks.
    if numberOfColorBands > 1
    
    ca = mat2cell(gr, blockVectorR, blockVectorC, numberOfColorBands); 
    else
    ca = mat2cell(gr, blockVectorR, blockVectorC);
    end
    % Display all the blocks.
    plotIndex = 1;
    numPlotsR = size(ca, 1);
    numPlotsC = size(ca, 2);
    for r = 1 : numPlotsR
    for c = 1 : numPlotsC
    for k = 1:length(ca)
    baseFileName = ca(k).name;
    I = imread(baseFileName);
    fprintf('plotindex = %d, c=%d, r=%d\n', plotIndex, c, r, I);
    subplot(numPlotsR, numPlotsC, plotIndex);
    rgbBlock = ca{r,c,I};
    imshow(rgbBlock);
    GLCM  = graycomatrix(rgbBlock,'Offset', [0 1;-1 1;-1 0;-1  -1],'NumLevel', 8,'Symmetric',true); 
    stats = GLCM_Features1(GLCM,'Contrast, Correlation, Energy, Homogeneity');
      h(k,1)= stats.contr(1);%contrast    
      h(k,2)= stats.corrp(1);%Correlation
      h(k,3)= stats.homom(1);%Homogeneity 
      h(k,4)= stats.energ(1);%Energy

      [rowsB, columnsB, numberOfColorBandsB] = size(rgbBlock);
      caption = sprintf('Block #%xzd of %d\n%d rows by %d columns', ...
        plotIndex, numPlotsR*numPlotsC, rowsB,  columnsB);
      title(caption);
      drawnow;
      plotIndex = plotIndex + 1;

          end
          end
          end

could you help me please for this ?

i realy need your help, please.

[Image Analyst]

ca is a cell array, not the result of a call to dir(). So why do you try to get the baseFileName = ca(k).name in the loop? It makes no sense. Somehow you took the FAQ code http://matlab.wikia.com/wiki/Split_image_into_blocks and altered it in a nonsensical way such that it no longer works. You do not need the base file name in the loop. In fact you do not need the look over k at all. Just follow the original code in the FAQ and put your call to graycomatrix inside the inner loop.

[Ade Aulya]

what should I do ?

[Image Analyst]

Get rid of that line, and the imread() line below it.

[Ade Aulya]

thank you so much for your help, sir. i'm sorry for my mistake. I just know now.

it's already run. but the problem is, how can I extract the glcm value from every single tile of an image, sir ? because when I run that code. it only extract the glcm of an image. And I need the glcm value not from an image but from each tile of the image. those glcm from each tile i will use for cluster coocurence matrix (CCM feature).

could u help me please, sir ?

[Image Analyst]

Why do you think it's doing the whole image? You're passing in rgbBlock, not the whole image so it's only doing that one block.

[Ade Aulya]

so, how can I read it to get glcm of each tiles from whole image, sir ?

[Image Analyst]

You get the matrix with this line of code:

GLCM  = graycomatrix(rgbBlock,'Offset', [0 1;-1 1;-1 0;-1  -1],'NumLevel', 8,'Symmetric',true);

It is the GLCM for just one tile.

[Ade Aulya]

yes sir. so if I want to get glcm for all tiles. what should I do, sir ? im sorry I still don't get the idea.

[Walter Roberson]

glcm_output = blockproc(YourRGBMatrix, [blockSizeR, blockSizeC], @(block_struct) graycomatrix(block_struct.data(:,:,1), 'Offset', [0 1;-1 1;-1 0;-1  -1], 'NumLevel', 8, 'Symmetric',true), 'TrimBorder', false);

glcm_output will be a 2D matrix that is all of the glcm results concatenated together. So if the output of glcm is 8 x 8, and there are (say) 39 groups of 100 pixels vertically, and 48 groups of 100 pixels horizontally (say the images was 384 x 480), the output would be (39*8) x (48*8) which would be 312 x 384.

Only the first color plane will be examined in this code, because glcm is only for grayscale not for color. Instead of block_struct.data(:,:,1) you could use rgb2gray(block_struct.data) to convert to grayscale... but it would be more efficient to instead

glcm_output = blockproc(rgb2gray(YourRGBMatrix), [blockSizeR, blockSizeC], @(block_struct) graycomatrix(block_struct.data, 'Offset', [0 1;-1 1;-1 0;-1  -1], 'NumLevel', 8, 'Symmetric',true), 'TrimBorder', false);

With a slight bit of extra work you could use a real function instead of an anonymous function, and have that real function call GLCM_Features1 on the output of the graycomatrix, and pick out the required stats and return them.

[Image Analyst]

I do have a version of blockproc() that uses a separate function, where you can do more complicated operations than in an anonymous function. It is attached.

There is also a function called graycoprops() that you may want instead of graycomatrix(). It gives you key features from the GLCM matrix.

Answer 3

[Ahmad]

hello. its work well on 256 by 256 image but i have 512 by 512 image. it gives the following error. Index exceeds matrix dimensions.

[Walter Roberson]

See the FAQ that Image Analyst posted.