mri.operators.linear.utils
mri.operators.linear.utils#
Functions used for the dictionary learning Compressed Sensing reconstruction.
- min_max_normalize(img)[source]#
Center and normalize the given array.
- Parameters
img (numpy.ndarray) –
- Returns
ndarray
- Return type
the center and normalized array.
- extract_patches_from_2d_images(img, patch_shape)[source]#
Return flattened patches from the 2d image.
- Parameters
img (numpy.ndarray of floats, the input 2d image) – patch_shape: tuple of int, shape of the patches
- Returns
patches – dim nb_patches*(patch.shape[0]*patch_shape[1])
- Return type
numpy.ndarray of floats, a 2d matrix with
- generate_flat_patches(images, patch_size, option='real')[source]#
Generate flat patches the list of images.
The generated images can be either the real, imaginary, complex or modulus of the images.
- Parameters
- Returns
flat_patches – The patches flat and concatained as a list
- Return type
list of np.ndarray as a GENERATOR
- learn_dictionary(flat_patches_subjects, nb_atoms=100, alpha=1, n_iter=1, fit_algorithm='lars', transform_algorithm='lasso_lars', batch_size=100, n_jobs=1, verbose=1)[source]#
Learn the dictionary from a training set.
- Parameters
flat_patches (generator of 1d array of flat patches (floats)) – a list per subject
nb_atoms (int,) – number of components of the dictionary (default=100)
alpha (float,) – regulation term (default=1)
n_iter (int) – number of iterations (default=1)
fit_algorithm ('lars') – for more details see MiniBatchDictionaryLearning from the sklearn library
transform_algorithm ('lasso_lars',) – for more details see MiniBatchDictionaryLearning from the sklearn library
batch_size (int (default 100),) – number of patches taken per iteration to fit the model
n_jobs (int defaul 6,) – number of cpu to run the learning
verbose (int default1,) – The level of verbosity
- Returns
dico
- Return type
MiniBatchDictionaryLearning object