improve multichannel handling and tests

skimage/transform/_warps.py

@@ -7,7 +7,7 @@
 
 
 def resize(image, output_shape, order=1, mode='constant', cval=0, clip=True,
-           preserve_range=False, multichannel=True):
+           preserve_range=False):
     """Resize image to match a certain size.
 
     Performs interpolation to up-size or down-size images. For down-sampling
@@ -48,8 +48,6 @@ def resize(image, output_shape, order=1, mode='constant', cval=0, clip=True,
     preserve_range : bool, optional
         Whether to keep the original range of values. Otherwise, the input
         image is converted according to the conventions of `img_as_float`.
-    multichannel : bool, optional
-        If True and ``image.ndim > 2``, treat last axis as channels.
 
     Examples
     --------
@@ -60,7 +58,6 @@ def resize(image, output_shape, order=1, mode='constant', cval=0, clip=True,
     (100, 100)
 
     """
-
     output_shape = np.asarray(output_shape)
     orig_shape = image.shape
     if len(orig_shape) < len(output_shape):
@@ -120,7 +117,7 @@ def resize(image, output_shape, order=1, mode='constant', cval=0, clip=True,
 
 
 def rescale(image, scale, order=1, mode='constant', cval=0, clip=True,
-            preserve_range=False):
+            preserve_range=False, multichannel=None):
     """Scale image by a certain factor.
 
     Performs interpolation to upscale or down-scale images. For down-sampling
@@ -159,6 +156,8 @@ def rescale(image, scale, order=1, mode='constant', cval=0, clip=True,
     preserve_range : bool, optional
         Whether to keep the original range of values. Otherwise, the input
         image is converted according to the conventions of `img_as_float`.
+    multichannel : bool, optional
+        If True, last axis will not be rescaled.
 
     Examples
     --------
@@ -171,11 +170,15 @@ def rescale(image, scale, order=1, mode='constant', cval=0, clip=True,
     (256, 256)
 
     """
+    if multichannel is None:
+        multichannel = False  # maintain previous default behavior
     scale = np.atleast_1d(scale)
     if len(scale) > 1 and len(scale) != image.ndim:
         raise ValueError("must supply a single scale or one value per axis.")
     orig_shape = np.asarray(image.shape)
     output_shape = np.round(scale * orig_shape).astype('i8')
+    if multichannel:  # don't scale channel dimension
+        output_shape[-1] = orig_shape[-1]
 
     return resize(image, output_shape, order=order, mode=mode, cval=cval,
                   clip=clip, preserve_range=preserve_range)

skimage/transform/pyramids.py

@@ -5,9 +5,20 @@
 from ..util import img_as_float
 
 
-def _smooth(image, sigma, mode, cval, multichannel=True):
-    """Return image with each channel smoothed by the Gaussian filter."""
+def _previous_multichannel_default(multichannel, ndim):
+    if multichannel is None:
+        # needed to maintain previous default behavior
+        if ndim == 3:
+            return True
+        else:
+            return False
+    else:
+        return multichannel
+
 
+def _smooth(image, sigma, mode, cval, multichannel=None):
+    """Return image with each channel smoothed by the Gaussian filter."""
+    multichannel = _previous_multichannel_default(multichannel, image.ndim)
     smoothed = np.empty(image.shape, dtype=np.double)
 
     # apply Gaussian filter to all channels independently
@@ -29,7 +40,7 @@ def _check_factor(factor):
 
 
 def pyramid_reduce(image, downscale=2, sigma=None, order=1,
-                   mode='reflect', cval=0, multichannel=True):
+                   mode='reflect', cval=0, multichannel=None):
     """Smooth and then downsample image.
 
     Parameters
@@ -63,13 +74,13 @@ def pyramid_reduce(image, downscale=2, sigma=None, order=1,
     .. [1] http://web.mit.edu/persci/people/adelson/pub_pdfs/pyramid83.pdf
 
     """
-
+    multichannel = _previous_multichannel_default(multichannel, image.ndim)
     _check_factor(downscale)
 
     image = img_as_float(image)
 
     out_shape = tuple([math.ceil(d / float(downscale)) for d in image.shape])
-    if multichannel and image.ndim > 2:
+    if multichannel:
         out_shape = out_shape[:-1]
 
     if sigma is None:
@@ -83,7 +94,7 @@ def pyramid_reduce(image, downscale=2, sigma=None, order=1,
 
 
 def pyramid_expand(image, upscale=2, sigma=None, order=1,
-                   mode='reflect', cval=0, multichannel=True):
+                   mode='reflect', cval=0, multichannel=None):
     """Upsample and then smooth image.
 
     Parameters
@@ -117,13 +128,13 @@ def pyramid_expand(image, upscale=2, sigma=None, order=1,
     .. [1] http://web.mit.edu/persci/people/adelson/pub_pdfs/pyramid83.pdf
 
     """
-
+    multichannel = _previous_multichannel_default(multichannel, image.ndim)
     _check_factor(upscale)
 
     image = img_as_float(image)
 
     out_shape = tuple([math.ceil(upscale * d) for d in image.shape])
-    if multichannel and image.ndim > 2:
+    if multichannel:
         out_shape = out_shape[:-1]
 
     if sigma is None:
@@ -138,7 +149,7 @@ def pyramid_expand(image, upscale=2, sigma=None, order=1,
 
 
 def pyramid_gaussian(image, max_layer=-1, downscale=2, sigma=None, order=1,
-                     mode='reflect', cval=0, multichannel=True):
+                     mode='reflect', cval=0, multichannel=None):
     """Yield images of the Gaussian pyramid formed by the input image.
 
     Recursively applies the `pyramid_reduce` function to the image, and yields
@@ -183,7 +194,6 @@ def pyramid_gaussian(image, max_layer=-1, downscale=2, sigma=None, order=1,
     .. [1] http://web.mit.edu/persci/people/adelson/pub_pdfs/pyramid83.pdf
 
     """
-
     _check_factor(downscale)
 
     # cast to float for consistent data type in pyramid
@@ -215,7 +225,7 @@ def pyramid_gaussian(image, max_layer=-1, downscale=2, sigma=None, order=1,
 
 
 def pyramid_laplacian(image, max_layer=-1, downscale=2, sigma=None, order=1,
-                      mode='reflect', cval=0, multichannel=True):
+                      mode='reflect', cval=0, multichannel=None):
     """Yield images of the laplacian pyramid formed by the input image.
 
     Each layer contains the difference between the downsampled and the
@@ -264,7 +274,7 @@ def pyramid_laplacian(image, max_layer=-1, downscale=2, sigma=None, order=1,
     .. [2] http://sepwww.stanford.edu/~morgan/texturematch/paper_html/node3.html
 
     """
-
+    multichannel = _previous_multichannel_default(multichannel, image.ndim)
     _check_factor(downscale)
 
     # cast to float for consistent data type in pyramid
@@ -288,12 +298,11 @@ def pyramid_laplacian(image, max_layer=-1, downscale=2, sigma=None, order=1,
         out_shape = tuple(
             [math.ceil(d / float(downscale)) for d in current_shape])
 
-        if multichannel and image.ndim > 2:
+        if multichannel:
             out_shape = out_shape[:-1]
 
         resized_image = resize(smoothed_image, out_shape,
-                               order=order, mode=mode, cval=cval,
-                               multichannel=multichannel)
+                               order=order, mode=mode, cval=cval)
         smoothed_image = _smooth(resized_image, sigma, mode, cval,
                                  multichannel)
 

skimage/transform/tests/test_warps.py

@@ -148,6 +148,35 @@ def test_rescale():
     assert_almost_equal(scaled, ref)
 
 
+def test_rescale_multichannel():
+    # 1D + channels
+    x = np.zeros((8, 3), dtype=np.double)
+    scaled = rescale(x, 2, order=0, multichannel=True)
+    assert_equal(scaled.shape, (16, 3))
+    # 2D
+    scaled = rescale(x, 2, order=0, multichannel=False)
+    assert_equal(scaled.shape, (16, 6))
+    # multichannel defaults to False
+    scaled = rescale(x, 2, order=0)
+    assert_equal(scaled.shape, (16, 6))
+
+    # 2D + channels
+    x = np.zeros((8, 8, 3), dtype=np.double)
+    scaled = rescale(x, 2, order=0, multichannel=True)
+    assert_equal(scaled.shape, (16, 16, 3))
+    # 3D
+    scaled = rescale(x, 2, order=0, multichannel=False)
+    assert_equal(scaled.shape, (16, 16, 6))
+
+    # 3D + channels
+    x = np.zeros((8, 8, 8, 3), dtype=np.double)
+    scaled = rescale(x, 2, order=0, multichannel=True)
+    assert_equal(scaled.shape, (16, 16, 16, 3))
+    # 4D
+    scaled = rescale(x, 2, order=0, multichannel=False)
+    assert_equal(scaled.shape, (16, 16, 16, 6))
+
+
 def test_resize2d():
     x = np.zeros((5, 5), dtype=np.double)
     x[1, 1] = 1
@@ -190,6 +219,17 @@ def test_resize2d_4d():
     assert_almost_equal(resized, ref)
 
 
+def test_resize_nd():
+    for dim in range(1, 6):
+        shape = 2 + np.arange(dim) * 2
+        x = np.ones(shape)
+        out_shape = np.asarray(shape) * 1.5
+        resized = resize(x, out_shape, order=0, mode='reflect')
+        expected_shape = 1.5 * shape
+        assert_equal(resized.shape, expected_shape)
+        assert np.all(resized == 1)
+
+
 def test_resize3d_bilinear():
     # bilinear 3rd dimension
     x = np.zeros((5, 5, 2), dtype=np.double)