vc2_bit_widths.pattern_generation: Heuristic test pattern generation

The routines in this module implement heuristcs for producing test patterns for VC-2 filters which produce near-maximum magnitude values in a target intermediate or final value.

Due to non-linearities in VC-2’s filters (i.e. rounding and quantisation), the test patterns generated are not true worst-case signals but rather a ‘best effort’ to get close to the worst case. Analysis test patterns will tend to be very close to worst-case signals while synthesis signals are likely to be modest under-estimates. Nevertheless, these signals often reach values well above real pictures and noise.

Test pattern generators

Analysis

make_analysis_maximising_pattern(input_array, target_array, tx, ty)

Create a test pattern which maximises a value within an intermediate/final output of an analysis filter.

Note

In lossless coding modes, test patterns which maximise a given value in the encoder also maximise the corresponding value in the decoder. Consequently this function may also be used to (indirectly) produce lossless decoder test patterns.

Warning

The returned test pattern is designed to maximise a real-valued implementation of the target filter. Though it is likely that this signal also maximises integer-based implementations (such as those used by VC-2) it is not guaranteed.

Parameters

input_arraySymbolArray

The input array to the analysis filter.

target_arrayInfiniteArray

An intermediate or final output array produced by the analysis filter within which a value should be maximised.

tx, tyint

The coordinates of the target value within target_array which is to be maximised.

Returns

test_pattern_specificationTestPatternSpecification

Synthesis

make_synthesis_maximising_pattern(analysis_input_array, analysis_transform_coeff_arrays, synthesis_target_array, synthesis_output_array, tx, ty)

Create a test pattern which, after lossy encoding, is likely to maximise an intermediate/final value of the synthesis filter.

Warning

Because (heavy) lossy VC-2 encoding is a non-linear process, finding encoder inputs which maximise the decoder output is not feasible in general. This function uses a simple heuristic (see Test pattern generation) to attempt to achieve this goal but cannot provide any guarantees about the extent to which it succeeds.

Parameters

analysis_input_arraySymbolArray

The input array to a compatible analysis filter for the synthesis filter whose values will be maximised.

analysis_transform_coeff_arrays{level: {orient: InfiniteArray, …}}

The final output arrays from a compatible analysis filter for the synthesis filter whose values will be maximised. These should be provided as a nested dictionary of the form produced by analysis_transform().

synthesis_target_arrayInfiniteArray

The intermediate or final output array produced by the synthesis filter within which a value should be maximised.

synthesis_output_arrayInfiniteArray

The output array for the synthesis filter.

tx, tyint

The index of the value within the synthesis_target_array which is to be maximised.

Returns

test_pattern_specificationTestPatternSpecification