Image quality and dimension

Hello here i attached my processed image using convn, the problem is when I process one image and save it to my local machine its perfect, but if I process 5000 images using same pipeline and save that 5000 images, the image shape and structures are collapsed I dont know why, I attached my code and processed images, if anybody know please try to help.

This is my single processed image.

If i use 5000 image in the same pipeline the result be like this

Both are same image filtered using same pipeline.

Here is my code

infolder = 'C:\Users\91894\Desktop\matlab\Dataset\images_001\images'   
outfolder = 'C:\Users\91894\Desktop\matlab\Dataset\images_001\processed image'   
imgFiles = dir([infolder,filesep,'\*.png']) ; 
N = length(imgFiles) ;   
refback = zeros(25, 25);
for i = 1:N
    thisFile = [infolder,filesep,imgFiles(i).name] ;  
    [filepath,name,ext] = fileparts(thisFile) ;   
    outFile = [outfolder,filesep,[name,ext]] ;  
    I = imread(thisFile) ;  
    J = imresize(I,[256 256]);
    for f = 1:4
        SpatDog = fspecial('gaussian',25,0.732) - fspecial('gaussian',25,0.785);
        FreqDog = fftshift(fft2(fftshift(SpatDog)));
        Y = abs(FreqDog)
        refback = refback + im2double(Y);
    end
    refback = (1/4) * refback;
    Y = fftshift(ifft2(fftshift(refback)));
    conv = convn(J,Y);
    imwrite(conv,outFile) ;   
end

I need to write the image to my laptop as its shown in the first image

Answer 1

[Matt J]

Well, I don't know if this is intentional, but every new image in the loop gets filtered by a different kernel because refback never gets reset to zero and just keeps accumulating in the line,

refback = refback + im2double(Y);

If the filter kernel is supposed to be constant, the creation of Y should be pulled out of the loop, as below. Also, since Y is generated using FFTs, it is recommendable to ensure that Y is real-valued.

infolder = 'C:\Users\91894\Desktop\matlab\Dataset\images_001\images'   
outfolder = 'C:\Users\91894\Desktop\matlab\Dataset\images_001\processed image'   
imgFiles = dir([infolder,filesep,'\*.png']) ; 
N = length(imgFiles) ;   
refback = zeros(25, 25);

    for f = 1:4
        SpatDog = fspecial('gaussian',25,0.732) - fspecial('gaussian',25,0.785);
        FreqDog = fftshift(fft2(fftshift(SpatDog)));
        Y = abs(FreqDog)
        refback = refback + im2double(Y);
    end
    refback = (1/4) * refback;
    Y = fftshift(ifft2(fftshift(refback)));
    
    Y=real(Y);

for i = 1:N
    thisFile = [infolder,filesep,imgFiles(i).name] ;  
    [filepath,name,ext] = fileparts(thisFile) ;   
    outFile = [outfolder,filesep,[name,ext]] ;  
    I = imread(thisFile) ;  
    J = imresize(I,[256 256]);

    conv = convn(J,Y);
    imwrite(conv,outFile) ;   
end

[Aravind Prabhu Gopala Krishnan]

Thank you very much it worked great.