Implement support for Y'CbCr matrices generation.

[KelSolaar]

@nick-shaw and I talked about implementing a new colour.RGB_to_YCbCr_matrix definition that would generate the matrices used by the various Y'CbCr flavours, it would eventually be used by the current colour.RGB_to_YCbCr and colour.YCbCr_to_RGB definitions.

It would be beneficial for generating the matrices for usage in other applications and also probably faster.

[KelSolaar]

Adding some related code for future reference, it misses the offset though:

def compute_YCbCr_matrix(weights, is_legal=False, bit_depth=8):
    Kr, Kb = weights
    Y_min, Y_max, C_min, C_max = colour.models.rgb.ycbcr.YCbCr_ranges(bit_depth, is_legal, False)
    Y = np.array([Kr, (1 - Kr - Kb), Kb])
    Cb = 0.5 * (np.array([0, 0, 1]) - Y) / (1 - Kb)
    Cr = 0.5 * (np.array([1, 0, 0]) - Y) / (1 - Kr)
    Y *= Y_max - Y_min
    Cb *= C_max - C_min
    Cr *= C_max - C_min
    return np.linalg.inv(np.vstack([Y, Cb, Cr]))

[nick-shaw]

Should that code not also include the is_int parameter?

Would it be useful also to create a corresponding compute_YCbCr_offsets() function, even if that was relatively trivial? It would presumably be something like:

def compute_YCbCr_offsets(weights, is_legal=False, bit_depth=8, is_int=False):
    Y_min, Y_max, C_min, C_max = colour.models.rgb.ycbcr.YCbCr_ranges(bit_depth, is_legal, is_int)
    Y_offset = Y_min
    C_offset = (C_min + C_max) / 2
    return np.array([Y_min, C_offset, C_offset])

Then you could calculate the matrix and offsets and do:

 YCbCr = colour.utilities.dot_vector(matrix, RGB) + offsets

[KelSolaar]

Absolutely! It is not complete at all, I just pasted it so that I don't have to scroll for 10 hours in Slack :)

[remia]

 YCbCr = colour.utilities.dot_vector(matrix, RGB) + offsets

Maybe propose an option to have the offset right into a 3x4 or 4x4 matrix as well ?

[KelSolaar]

This raises the question as to where the offsets would go in the 4x4 matrix.