Usage (GLCM Cross)¶

Given that ar is an np.ndarray.

OOP GLCM Cross Transform

>>> from glcm_cupy import GLCMCross
>>> g = GLCMCross(...).run(ar)

Functional GLCM Cross Transform

>>> from glcm_cupy import glcm_cross
>>> g = glcm_cross(ar, ...)

I/O¶

For a np.ndarray or cp.ndarray input, the algorithm will output a np.ndarray or cp.ndarray respectively.

Input Array Requirements¶

Shape: \(B\times H\times W\times C\) or \(H\times W\times C\)
Non-negative, integer array
Must be large enough for radius
Must have at least 2 channels to cross

\[\begin{split}B: \text{Batches}\\ H: \text{Height}\\ W: \text{Width}\\ C: \text{Channel}\end{split}\]

Output Array¶

Shape: \(B\times H^*\times W^*\times C_{combo}\times F\) or \(H\times W\times C_{combo}\times F\)
0 to 1 if normalize_features, else it depends on feature.

\[\begin{split}H^*: H - \text{Radius} \times 2\\ W^*: W - \text{Radius} \times 2\\ C_{combo}: {C \choose 2} = \text{Number of Pair Combinations}\\ F: \text{Features}\end{split}\]

Combinations¶

The pair combination order is dependent on itertools.combinations.

For example, for an image with 4 channels:

>>> from itertools import combinations
>>> n_channels = 4
>>> pair_combinations = combinations(range(n_channels), 2)
>>> print(pair_combinations)
[(0, 1), (0, 2), (0, 3), (1, 2), (1, 3), (2, 3)]

We see that the 1st channel crosses Channel 0 and Channel 1.

Arguments¶

Argument	Description	Default
features	List of `GLCM.Features` for GLCM to use	`(Features.HOMOGENEITY, Features....)`
ix_combos	List of pair combinations to use	`None` (All combinations)
radius	Radius of GLCM Windows	`2`
bin_from	Maximum Value + 1 of the array.	`256`
bin_to	Maximum Value + 1 of the array	`256`
normalized_features	Whether to scale features to `[0, 1]`	`True`
verbose	Whether `tqdm` outputs progress	`True`
max_partition_size1	No. of windows parsed per GLCM Matrix	`10000`
max_threads1	No. of threads per CUDA block	`512`