vc2_quantisation_matrices Python module usage

The vc2_quantisation_matrices may be used to create quantisation matrices which normalise noise power between transform bands. This may be done using the following function:

derive_quantisation_matrix(wavelet_index, wavelet_index_ho, dwt_depth, dwt_depth_ho)

Derive a noise-power normalising quantisation matrix for the specified wavelet transform. This quantisation matrix will seek to cause quantisation noise-power to be spread evenly over all wavelet levels and bands.

Parameters

wavelet_index, wavelet_index_hovc2_data_tables.WaveletFilters

The vertical and horizontal wavelet filter indices respectively.

dwt_depth, dwt_depth_ho

The wavelet transform depth (2D depth and horizontal only depth respectively).

Returns

{level: {band: int, …}, …}

A quantisation matrix, as laid out by the quant_matrix VC-2 pseudocode function (12.4.5.3) and in vc2_data_tables.QUANTISATION_MATRICES.

As an example, the quantisation matrix for a 4-level 2D LeGall transform may be found as follows:

>>> from vc2_data_tables import WaveletFilters
>>> from vc2_quantisation_matrices import derive_quantisation_matrix

>>> matrix = derive_quantisation_matrix(
...     wavelet_index=WaveletFilters.le_gall_5_3,     # 1
...     wavelet_index_ho=WaveletFilters.le_gall_5_3,  # 1
...     dwt_depth=4,
...     dwt_depth_ho=0,
... )

>>> from pprint import pprint
>>> pprint(matrix)
{0: {'LL': 4},
 1: {'HH': 0, 'HL': 2, 'LH': 2},
 2: {'HH': 2, 'HL': 4, 'LH': 4},
 3: {'HH': 3, 'HL': 5, 'LH': 5},
 4: {'HH': 5, 'HL': 7, 'LH': 7}}

Tip

The vc2_data_tables module, used here, provides named constants for all of the VC-2 filter types.

The returned nested dictionary structure matches the layout used by the quant_matrix() VC-2 pseudocode function (12.4.5.3). For each transform level, the outer dictionary contains an entry with an inner dictionary giving the quantisation index offset for each orientation. Orientations are labelled using the strings ‘L’, ‘H’, ‘LL’ ‘LH’, ‘HL’, and ‘HH’.