diff --git a/CHANGELOG.rst b/CHANGELOG.rst
index 31b62178cd2d711fae19645e813e666be314e467..b6dad1cea394714ff0a04fe33fd78a207858b388 100644
--- a/CHANGELOG.rst
+++ b/CHANGELOG.rst
@@ -21,6 +21,8 @@ Changed
* Minimum required version of ``nibabel`` is now 2.3.
* The :class:`.Image` class now fully delegates to ``nibabel`` for managing
file handles.
+* The :meth:`.Image.resample` method now supports images with more than three
+ dimensions.
Removed
@@ -38,6 +40,7 @@ Fixed
* Make sure that FEAT ``Cluster`` objects (created by the
:func:`.loadClusterResults` function) contain ``p`` and ``logp`` attributes,
even when cluster thresholding was not used.
+* Fix to the :class:`.ImageWrapper` regarding complex data types.
1.13.0 (Thursday 22nd November 2018)
diff --git a/fsl/data/image.py b/fsl/data/image.py
index a1e695b59bd81a36a845d3f64167efbf6222ecc5..24e129c8bfc1ecc42d56d5c77d241e3912c5bd67 100644
--- a/fsl/data/image.py
+++ b/fsl/data/image.py
@@ -1173,7 +1173,7 @@ class Image(Nifti):
order=1,
smooth=True):
"""Returns a copy of the data in this ``Image``, resampled to the
- specified ``shape``.
+ specified ``newShape``.
:arg newShape: Desired shape. May containg floating point values,
in which case the resampled image will have shape
@@ -1182,7 +1182,7 @@ class Image(Nifti):
:arg sliceobj: Slice into this ``Image``. If ``None``, the whole
image is resampled, and it is assumed that it has the
- same number of dimensions as ``shape``. A
+ same number of dimensions as ``newShape``. A
:exc:`ValueError` is raised if this is not the case.
:arg dtype: ``numpy`` data type of the resampled data. If ``None``,
@@ -1201,12 +1201,12 @@ class Image(Nifti):
:returns: A tuple containing:
- - A ``numpy`` array of shape ``shape``, containing an
- interpolated copy of the data in this ``Image``.
+ - A ``numpy`` array of shape ``newShape``, containing
+ an interpolated copy of the data in this ``Image``.
- A ``numpy`` array of shape ``(4, 4)``, containing the
- adjusted voxel-to-world transformation for the resampled
- data.
+ adjusted voxel-to-world transformation for the spatial
+ dimensions of the resampled data.
"""
if sliceobj is None: sliceobj = slice(None)
@@ -1225,7 +1225,7 @@ class Image(Nifti):
ratio = oldShape / newShape
newShape = np.array(np.round(newShape), dtype=np.int)
- scale = transform.scaleOffsetXform(ratio, 0)
+ scale = np.diag(ratio)
# If interpolating and smoothing, we apply a
# gaussian filter along axes with a resampling
@@ -1242,7 +1242,7 @@ class Image(Nifti):
data = ndimage.gaussian_filter(data, sigma)
data = ndimage.affine_transform(data,
- scale[:3, :3],
+ scale,
output_shape=newShape,
order=order)
@@ -1250,6 +1250,7 @@ class Image(Nifti):
# puts the resampled image into the
# same world coordinate system as this
# image.
+ scale = transform.scaleOffsetXform(ratio[:3], 0)
xform = transform.concat(self.voxToWorldMat, scale)
else:
xform = self.voxToWorldMat
diff --git a/fsl/data/imagewrapper.py b/fsl/data/imagewrapper.py
index b73cb879aaf88ffa1e742477b3625fd7da91a590..2b49e4af594268f49a48a5dc280c323bc45d8eda 100644
--- a/fsl/data/imagewrapper.py
+++ b/fsl/data/imagewrapper.py
@@ -301,7 +301,14 @@ class ImageWrapper(notifier.Notifier):
# Internally, we calculate and store the
# data range for each volume/slice/vector
- self.__volRanges = np.zeros((nvols, 2), dtype=np.float32)
+ #
+ # We use nan as a placeholder, so the
+ # dtype must be non-integral
+ dtype = self.__image.get_data_dtype()
+ if np.issubdtype(dtype, np.integer):
+ dtype = np.float32
+ self.__volRanges = np.zeros((nvols, 2),
+ dtype=dtype)
self.__coverage[ :] = np.nan
self.__volRanges[:] = np.nan
diff --git a/tests/test_image.py b/tests/test_image.py
index 2786cc501e3a3eee2bb387cd78c33fd303d79b27..3290ed3619e6d13c713977cf851dd80bd37843c8 100644
--- a/tests/test_image.py
+++ b/tests/test_image.py
@@ -1078,6 +1078,12 @@ def test_image_resample(seed):
make_random_image(fname, shape)
img = fslimage.Image(fname, mmap=False)
+ # bad shape
+ with pytest.raises(ValueError):
+ img.resample((10, 10))
+ with pytest.raises(ValueError):
+ img.resample((10, 10, 10, 10))
+
# resampling to the same shape should be a no-op
samei, samex = img.resample(shape)
assert np.all(samei == img[:])
@@ -1134,18 +1140,50 @@ def test_image_resample(seed):
assert np.all(np.isclose(resvals, origvals))
- # Test a 4D image
+ del img
+ img = None
+
+
+def test_image_resample_4d(seed):
+
+ fname = 'test.nii.gz'
+
+ with tempdir():
+
make_random_image(fname, (10, 10, 10, 10))
+
+ # resample one volume
img = fslimage.Image(fname)
slc = (slice(None), slice(None), slice(None), 3)
-
resampled = img.resample(img.shape[:3], slc)[0]
assert np.all(resampled == img[..., 3])
+ # resample up
resampled = img.resample((15, 15, 15), slc)[0]
assert tuple(resampled.shape) == (15, 15, 15)
+
+ # resample down
+ resampled = img.resample((5, 5, 5), slc)[0]
+ assert tuple(resampled.shape) == (5, 5, 5)
+
+ # resample the entire image
+ resampled = img.resample((15, 15, 15, 10), None)[0]
+ assert tuple(resampled.shape) == (15, 15, 15, 10)
+
+ resampled = img.resample((5, 5, 5, 10), None)[0]
+ assert tuple(resampled.shape) == (5, 5, 5, 10)
+
+ # resample along the fourth dim
+ resampled = img.resample((15, 15, 15, 15), None)[0]
+ assert tuple(resampled.shape) == (15, 15, 15, 15)
+
+ resampled = img.resample((5, 5, 5, 15), None)[0]
+ assert tuple(resampled.shape) == (5, 5, 5, 15)
+
del img
+ del resampled
img = None
+ resampled = None
def test_Image_init_xform_nifti1(): _test_Image_init_xform(1)