Can anyone please help me with this code, this should correct a skewed image using principal component analysis. The output image somehow corrected but it is not perfect because it still have a small skew

 clear;
    img = imread('test3.jpg'); 
    if length(size(img))>2
    img = rgb2gray(img);
    end 
%applying sobel edge detector in the horizontal direction
    fx = [-1 0 1;-1 0 1;-1 0 1];
    Ix = filter2(fx,img);
% applying sobel edge detector in the vertical direction
    fy = [1 1 1;0 0 0;-1 -1 -1];
    Iy = filter2(fy,img); 

    Ix2 = Ix.^2;
    Iy2 = Iy.^2;
    Ixy = Ix.*Iy;
    clear Ix;
    clear Iy;

    %applying gaussian filter on the computed value
    h= fspecial('gaussian',[7 7],2); 
    Ix2 = filter2(h,Ix2);
    Iy2 = filter2(h,Iy2);
    Ixy = filter2(h,Ixy);
    height = size(img,1);
    width = size(img,2);
    result = zeros(height,width); 
    R = zeros(height,width);

    ModifyCorner = 100; %Value for number of corner plots that will be detected

    GetCornerPoints = corner(img,ModifyCorner) 
    imshow(img); hold on;

   %plotting of corner points
    plot(GetCornerPoints(:,1), GetCornerPoints(:,2), 'g*');

 newpca= pca(GetCornerPoints) %getting the pca of the cornerpoints plotted

 PcaCol1= sum(newpca(:,1)) %the sum of Pca column 1
 PcaCol2= sum(newpca(:,2)) %the sum of Pca column 2

      NewHigh = max(newpca(:)) %getting the highest pca value

%this if-else statement is used to determine the ange of rotation for the image
  if PcaCol1>PcaCol2
      %getting the angle of rotation
  GetTempo=acos(abs(NewHigh))*(180/3.14)

    GetAngle=-abs(GetTempo) %converts the angle into a negative value
    RotateImage = imrotate(img, GetAngle,'bilinear'); %rotation of the image
  imshow(img)
  figure, imshow(RotateImage)
    else 
        %getting the angle of rotation
         GetTempo=acos(abs(NewHigh))*(180/3.14)

   GetAngle=abs(GetTempo)
   RotateImage = imrotate(img, GetAngle,'bilinear'); %rotation of the image
  imshow(img)
  figure, imshow(RotateImage)

    end

[Enrick Dela Cruz]

This is the input image

[Enrick Dela Cruz]

This is the Output image

Answer 1

[Image Analyst]

Try using hough() to identify lines and find the angle that most of them are at. See the "gantry crane" demo in the help for hough.

By the way, this is not skew, which would be more like a shear. It's more like just an ordinary rotation.

It might not be necessary to rotate the image to find the license plate letters and numbers, which is what I think you're after. Are you 100% sure it is?

[Enrick Dela Cruz]

Actually we need to prove which algorithm is the best in correcting skew, we are comparing the results of pca, hough transform and radon. any opinion sir?

[Enrick Dela Cruz]

also do you think my code of pca in correcting skew is fine or do it still need some changes to get a better result?

[Image Analyst]

They all make different assumptions, or I guess I should say that each method will work best for images that have certain kinds of content in them. So try them all against a known image that you've skewed and compute ssim, psnr, or mse.

[Enrick Dela Cruz]

Thank you sir, i will still try to fix the code to have a good output.