Image rotation based on PCA in Matlab

I'm working with an image of an array where we on one side has added an extra row, this in order to make the data scatter greater in one direction. Now I want to rotate the image I my idea is to use PCA. I started off with preprocess the image, segment it and then I only keep the array part for the PCA.

[I, J] = find(ims);
mat = [I, J];
covariance = cov(mat);
[eigvect, eigval] = eig(covariance);
% Chose the eigenvector with highest eigenvalue
if (max(eigval(:,1)) > max(eigval(:,2)))
    v = eigvect(:,1);
else
    v = eigvect(:,2);
end

v1 = v./norm(v,1);
v2 = [0,1];
%Calculate angle
angle = acosd( dot(v1,v2) / (v1 * norm(v2)) );

My questions are:

1. Is it better to use the original grayscale image and perform the PCA on it rather than the binary image?


2. Am I wrong using cov instead of pca? My chose is based on the fact that I will eventually change to openCV and implement everything in a smartphone application so I want to use methods that I can find in openCV too.


3. Why do I get two angles when I only want the angle between the eigenvector and the vertical vector [0,1]?


4. Do you think I should add an extra element to my array in order to increase the difference in how the data is spread?

Thanks for your time :)