# -*- coding: utf-8 -*-
##########################################################################
# pySAP - Copyright (C) CEA, 2017 - 2018
# Distributed under the terms of the CeCILL-B license, as published by
# the CEA-CNRS-INRIA. Refer to the LICENSE file or to
# http://www.cecill.info/licences/Licence_CeCILL-B_V1-en.html
# for details.
##########################################################################
"""
This module contains all the function to flatten properly the cube from a ISAP
.fits raw data.
"""
import numpy as np
###
# HELPERS
# GETTERS
[docs]def get_hbl(A):
"""Return the half-bottom-left of the given array.
"""
nx, _ = A.shape
li = int(nx / 2)
return A[li:, :li]
[docs]def get_hbr(A):
"""Return the half-bottom-right of the given array.
"""
nx, _ = A.shape
li = int(nx / 2)
return A[li:, li:]
[docs]def get_htl(A):
"""Return the half-top-left of the given array.
"""
nx, _ = A.shape
li = int(nx / 2)
return A[:li, :li]
[docs]def get_htr(A):
"""Return the half-top-right of the given array.
"""
nx, _ = A.shape
li = int(nx / 2)
return A[:li, li:]
[docs]def get_hr(A):
"""Return the half-right of the given array.
"""
nx, _ = A.shape
li = int(nx / 2)
return A[:, li:]
[docs]def get_hl(A):
"""Return the half-left of the given array.
"""
nx, _ = A.shape
li = int(nx / 2)
return A[:, :li]
[docs]def get_hb(A):
"""Return the half-bottom of the given array.
"""
nx, _ = A.shape
li = int(nx / 2)
return A[li:, :]
[docs]def get_ht(A):
"""Return the half-top of the given array.
"""
nx, _ = A.shape
li = int(nx / 2)
return A[:li, :]
# SETTERS
[docs]def set_hbl(A, a):
"""Return the half-bottom-left of the given array.
"""
nx, _ = A.shape
li = int(nx / 2)
A[li:, :li] = a
[docs]def set_hbr(A, a):
"""Return the half-bottom-right of the given array.
"""
nx, _ = A.shape
li = int(nx / 2)
A[li:, li:] = a
[docs]def set_htl(A, a):
"""Return the half-top-left of the given array.
"""
nx, _ = A.shape
li = int(nx / 2)
A[:li, :li] = a
[docs]def set_htr(A, a):
"""Return the half-top-right of the given array.
"""
nx, _ = A.shape
li = int(nx / 2)
A[:li, li:] = a
[docs]def set_hr(A, a):
"""Return the half-right of the given array.
"""
nx, _ = A.shape
li = int(nx / 2)
A[:, li:] = a
[docs]def set_hl(A, a):
"""Return the half-left of the given array.
"""
nx, _ = A.shape
li = int(nx / 2)
A[:, :li] = a
[docs]def set_hb(A, a):
"""Return the half-bottom of the given array.
"""
nx, _ = A.shape
li = int(nx / 2)
A[li:, :] = a
[docs]def set_ht(A, a):
"""Return the half-top of the given array.
"""
nx, _ = A.shape
li = int(nx / 2)
A[:li, :] = a
###
# FLATTEN
[docs]def flatten_undecimated_n_bands(cube, trf):
"""Flatten the decomposition coefficients from a cube to a vector.
``flatten_undecimated_n_bands`` concerns the cube where each layer
correspond to a undecimated band. We can have multiple bands per scale,
which lead to ``nb_scale * nb_band_per_scale`` for one dimension.
Parameters
----------
cube: numpy.ndarray
The cube that contains the decomposition coefficients
Returns
-------
numpy.ndarray
The flattened cube
"""
return np.copy(cube.flatten())
[docs]def flatten_decimated_1_bands(cube, trf):
"""Flatten the decomposition coefficients from a cube to a vector.
``flatten_decimated_1_bands`` concerns the cube where it's actually a
2d-array like the classic wavelet 2d-transform of 1 bands. It has the same
formating than the 3 bands but the ``v`` and ``h`` bands or set to ``0``.
Parameters
----------
cube: numpy.ndarray
The cube that contains the decomposition coefficients
Returns
-------
numpy.ndarray
The flattened cube
"""
pieces = []
for i in range(trf.nb_scale - 1):
pieces.append(get_htl(cube).flatten())
cube = get_hbr(cube)
pieces.append(cube.flatten()) # get approx
return np.concatenate(pieces)
[docs]def flatten_decimated_3_bands(cube, trf):
"""Flatten the decomposition coefficients from a cube to a vector.
``flatten_decimated_3_bands`` concerns the cube where it's actually
a 2d-array like the classic wavelet 2d-transform of 3 bands.
Parameters
----------
cube: numpy.ndarray
The cube that containes the decomposition coefficients
Returns
-------
numpy.ndarray
The flattened cube
"""
pieces = []
for i in range(trf.nb_scale - 1):
pieces.append(get_htr(cube).flatten())
pieces.append(get_hbr(cube).flatten())
pieces.append(get_hbl(cube).flatten())
cube = get_htl(cube)
pieces.append(cube.flatten()) # get approx
return np.concatenate(pieces)
[docs]def flatten_vector(cube, trf):
"""Flatten the decomposition coefficients from a cube to a vector.
``flatten_vector`` concerns the ``curvelet-cube`` where it's already
a vector.
Parameters
----------
cube: numpy.ndarray
The cube that containes the decomposition coefficients
Returns
-------
numpy.ndarray
The flattened cube
"""
metadata_len = 1 + trf.nb_scale + 2 * trf.nb_band_per_scale.sum()
data = np.zeros(len(cube) - metadata_len)
cube_padd = 1 + trf.nb_scale + 2
data_padd = 0
for ks in range(trf.nb_scale):
for kb in range(trf.nb_band_per_scale[ks]):
tmp = cube[cube_padd:cube_padd + trf.bands_lengths[ks, kb]]
data[data_padd:data_padd + trf.bands_lengths[ks, kb]] = tmp
Nx = trf.bands_shapes[ks][kb][0]
Ny = trf.bands_shapes[ks][kb][1]
cube_padd += (Nx * Ny + 2)
data_padd += (Nx * Ny)
return data
[docs]def flatten_decimated_feauveau(cube, trf):
"""Flatten decomposition coefficients from a cube to a vector.
``flatten_decimated_feauveau`` concern the cube where it's the Feauveau
decimated.
Parameters
----------
cube: numpy.ndarray
The cube that containes the decomposition coefficients
Returns
-------
numpy.ndarray
The flattened cube
"""
pieces = []
for i in range(trf.nb_scale - 1):
pieces.append(get_hbl(cube).flatten())
pieces.append(get_hr(cube).flatten())
cube = get_htl(cube)
pieces.append(cube.flatten()) # get approx
return np.concatenate(pieces)
###
# INFLATED
[docs]def inflated_undecimated_n_bands(trf):
"""Inflated the decomposition coefficients from a vector to a cube.
``inflated_undecimated_n_bands`` concerns the vector where each layer
correspond to a undecimated band. We can have multiple bands per scale,
which lead to ``nb_scale * nb_band_per_scale`` for one dimension.
Returns
-------
numpy.ndarray
The unflattened cube
"""
return np.copy(trf._analysis_data.reshape(trf._analysis_shape))
[docs]def inflated_decimated_1_bands(trf):
"""Inflated the decomposition coefficients from a vector to a cube.
``inflated_decimated_1_bands`` concerns the vector where it's actually a
2d-array like the classic wavelet 2d-transform of 1 bands. It has the
same formating than the 3 bands but the ``v`` and ``h`` bands or set to
``0``.
Returns
-------
numpy.ndarray
The unflattened cube
"""
cube = np.zeros(trf._analysis_shape, dtype=trf._analysis_data.dtype)
tmp = cube
for ks in range(trf.nb_scale - 1):
set_htl(tmp, trf[ks, 0])
if ks == ((trf.nb_scale - 1) - 1):
break
else:
tmp = get_hbr(tmp)
set_hbr(tmp, trf[trf.nb_scale - 1, 0]) # set approx
return cube
[docs]def inflated_decimated_3_bands(trf):
"""Inflated the decomposition coefficients from a vector to a cube.
``inflated_decimated_3_bands`` concern the vector where it's actually a
2d-array like the classic wavelet 2d-transform of 3 bands.
Returns
-------
numpy.ndarray
The unflattened cube
"""
cube = np.zeros(trf._analysis_shape, dtype=trf._analysis_data.dtype)
tmp = cube
for ks in range(trf.nb_scale - 1):
set_htr(tmp, trf[ks, 0])
set_hbr(tmp, trf[ks, 1])
set_hbl(tmp, trf[ks, 2])
if ks == ((trf.nb_scale - 1) - 1):
break
else:
tmp = get_htl(tmp)
set_htl(tmp, trf[trf.nb_scale - 1, 0]) # set approx
return cube
[docs]def inflated_vector(trf):
"""Inflated the decomposition coefficients from a vector to a cube.
``inflated_vector`` concerns the vector where it's encode a curvelet.
Returns
-------
numpy.ndarray
The unflattened cube
"""
metadata_len = 1 + trf.nb_scale + 2 * trf.nb_band_per_scale.sum()
cube = np.zeros(len(trf._analysis_data) + metadata_len)
cube[0] = trf.nb_scale
cube[1:1 + trf.nb_scale] = trf.nb_band_per_scale
cube_padd = 1 + trf.nb_scale
data_padd = 0
for ks in range(trf.nb_scale):
for kb in range(trf.nb_band_per_scale[ks]):
Nx = trf.bands_shapes[ks][kb][0]
Ny = trf.bands_shapes[ks][kb][1]
cube[cube_padd] = Nx
cube_padd += 1
cube[cube_padd] = Ny
cube_padd += 1
tmp = trf[ks, kb].flatten()
cube[cube_padd:cube_padd + trf.bands_lengths[ks, kb]] = tmp
cube_padd += (Nx * Ny)
data_padd += (Nx * Ny)
return cube
[docs]def inflated_decimated_feauveau(trf):
"""Inflated the decomposition coefficients from a vector to a cube.
``inflated_decimated_feauveau`` concerns the vector where it's the
Feauveau decimated.
Returns
-------
numpy.ndarray
The unflattened cube
"""
cube = np.zeros(trf._analysis_shape, dtype=trf._analysis_data.dtype)
tmp = cube
for ks in range(trf.nb_scale - 1):
set_hbl(tmp, trf[ks, 0])
set_hr(tmp, trf[ks, 1])
if ks == ((trf.nb_scale - 1) - 1):
break
else:
tmp = get_htl(tmp)
set_htl(tmp, trf[trf.nb_scale - 1, 0]) # set approx
return cube
###
# FORMATING FCTS INDEXES
FLATTENING_FCTS = [flatten_undecimated_n_bands,
flatten_decimated_1_bands,
flatten_decimated_3_bands,
flatten_vector,
flatten_decimated_feauveau]
INFLATING_FCTS = [inflated_undecimated_n_bands,
inflated_decimated_1_bands,
inflated_decimated_3_bands,
inflated_vector,
inflated_decimated_feauveau]
