Potential test case: Linear intensity matching

[Romain-Laine]

Here's a suggestion of a test case: perform intensity rescaling of one image to match the intensity distribution of another image, by linear rescaling. This can be done by simple explicit minimisation of the mean square error, which leads to this expression:

def linear_intensity_matching(imA: np.ndarray, imB: np.ndarray) -> np.ndarray:
    """
    Perform a linear intensity match between 2 images so that they are in 
    the same intensity range, assuming: imB ~ slope*imA + intercept
    Arguments:
    - imA: first image, the one that will be adjusted
    - imB: the second image, the one that sets the range to match
    Output:
    - a rescaled version of imA such that  
    """
    meanA = imA.mean()
    meanB = imB.mean()
    meanAB = (imA*imB).mean()
    meanA2 = (imA**2).mean()

    slope = (meanAB-meanA*meanB)/(meanA2-meanA*meanB)
    intercept = meanB - slope*meanA

    return slope*imA+intercept

The use case for instance is to compare images through MSE afterwards once they're intensity matched, or applying other metrics to compare similarity.