Merge pull request #39 from blowekamp/update_fft_initialize

Add optional parameters fft_based_translation_initialization

SimpleITK/utilities/fft.py

@@ -20,7 +20,12 @@
 
 
 def fft_based_translation_initialization(
-    fixed: sitk.Image, moving: sitk.Image
+    fixed: sitk.Image,
+    moving: sitk.Image,
+    *,
+    required_fraction_of_overlapping_pixels: float = 0.0,
+    initial_transform: sitk.Transform = None,
+    masked_pixel_value: float = None,
 ) -> sitk.TranslationTransform:
     """Perform fast Fourier transform based normalized correlation to find the translation which maximizes correlation
     between the images.
@@ -34,30 +39,55 @@ def fft_based_translation_initialization(
     :param fixed: A SimpleITK image object.
     :param moving: Another SimpleITK Image object, which will be resampled onto the grid of the fixed image if it is not
         congruent.
-    :return: A TranslationTransform mapping physical points from the fixed to the moving image.
+    :param required_fraction_of_overlapping_pixels: The required fraction of overlapping pixels between the fixed and
+        moving image. The value should be in the range of [0, 1]. If the value is 1, then the full overlap is required.
+    :param initial_transform: An initial transformation to be applied to the moving image by resampling before the
+        FFT registration. The returned transform will be of the initial_transform type with the translation updated.
+    :param masked_pixel_value: The value of input pixels to be ignored by correlation. If None, then the
+        FFTNormalizedCoorrelation will be used, otherwise the MaskedFFTNormalizedCorrelation will be used.
+    :return: A TranslationTransform (or the initial_transform tyype) mapping physical points from the fixed to the
+     moving image.
     """
 
     if (
-        moving.GetSpacing() != fixed.GetSpacing()
+        initial_transform is not None
+        or moving.GetSpacing() != fixed.GetSpacing()
         or moving.GetDirection() != fixed.GetDirection()
         or moving.GetOrigin() != fixed.GetOrigin()
     ):
         resampler = sitk.ResampleImageFilter()
         resampler.SetReferenceImage(fixed)
+
+        if initial_transform is not None:
+            resampler.SetTransform(initial_transform)
         moving = resampler.Execute(moving)
 
     sigma = fixed.GetSpacing()[0]
     pixel_type = sitk.sitkFloat32
 
-    fft_fixed = sitk.Cast(sitk.SmoothingRecursiveGaussian(fixed, sigma), pixel_type)
-    fft_moving = sitk.Cast(sitk.SmoothingRecursiveGaussian(moving, sigma), pixel_type)
-
-    out = sitk.FFTNormalizedCorrelation(fft_fixed, fft_moving)
-
-    out = sitk.SmoothingRecursiveGaussian(out)
-    cc = sitk.ConnectedComponent(sitk.RegionalMaxima(out, fullyConnected=True))
+    fixed = sitk.Cast(sitk.SmoothingRecursiveGaussian(fixed, sigma), pixel_type)
+    moving = sitk.Cast(sitk.SmoothingRecursiveGaussian(moving, sigma), pixel_type)
+
+    if masked_pixel_value is None:
+        xcorr = sitk.FFTNormalizedCorrelation(
+            fixed,
+            moving,
+            requiredFractionOfOverlappingPixels=required_fraction_of_overlapping_pixels,
+        )
+    else:
+        xcorr = sitk.MaskedFFTNormalizedCorrelation(
+            fixed,
+            moving,
+            sitk.Cast(fixed != masked_pixel_value, pixel_type),
+            sitk.Cast(moving != masked_pixel_value, pixel_type),
+            requiredFractionOfOverlappingPixels=required_fraction_of_overlapping_pixels,
+        )
+
+    xcorr = sitk.SmoothingRecursiveGaussian(xcorr, sigma)
+
+    cc = sitk.ConnectedComponent(sitk.RegionalMaxima(xcorr, fullyConnected=True))
     stats = sitk.LabelStatisticsImageFilter()
-    stats.Execute(out, cc)
+    stats.Execute(xcorr, cc)
     labels = sorted(stats.GetLabels(), key=lambda l: stats.GetMean(l))
 
     peak_bb = stats.GetBoundingBox(labels[-1])
@@ -67,13 +97,21 @@ def fft_based_translation_initialization(
         for min_idx, max_idx in zip(peak_bb[0::2], peak_bb[1::2])
     ]
 
-    peak_pt = out.TransformContinuousIndexToPhysicalPoint(peak_idx)
+    peak_pt = xcorr.TransformContinuousIndexToPhysicalPoint(peak_idx)
     peak_value = stats.GetMean(labels[-1])
 
-    center_pt = out.TransformContinuousIndexToPhysicalPoint(
-        [p / 2.0 for p in out.GetSize()]
+    center_pt = xcorr.TransformContinuousIndexToPhysicalPoint(
+        [p / 2.0 for p in xcorr.GetSize()]
     )
 
     translation = [c - p for c, p in zip(center_pt, peak_pt)]
+    if initial_transform is not None:
+        offset = initial_transform.TransformVector(translation, point=[0, 0])
+
+        tx_out = sitk.Transform(initial_transform).Downcast()
+        tx_out.SetTranslation(
+            [a + b for (a, b) in zip(initial_transform.GetTranslation(), offset)]
+        )
+        return tx_out
 
-    return sitk.TranslationTransform(out.GetDimension(), translation)
+    return sitk.TranslationTransform(xcorr.GetDimension(), translation)

requirements.txt

@@ -1,2 +1,2 @@
-SimpleITK>=2.0
+SimpleITK>=2.3.0
 numpy

test/test_utilities.py

@@ -47,6 +47,7 @@ def test_fft_initialization():
 
 
 def test_fft_initialization2():
+    """Testing with different spacing and origin to force resampling."""
     fixed_img = sitk.Image([1024, 512], sitk.sitkUInt8)
 
     fixed_img[510:520, 255:265] = 10
@@ -60,6 +61,44 @@ def test_fft_initialization2():
     assert tx.GetOffset() == (-85.0, -105.0)
 
 
+def test_fft_initialization3():
+    """Testing with required fraction of overlapping pixels."""
+    fixed_img = sitk.Image([1024, 512], sitk.sitkUInt8)
+    fixed_img[0:10, 0:20] = 10
+    fixed_img[510:520, 255:265] = 10
+
+    moving_img = sitk.Image([1024, 512], sitk.sitkUInt8)
+    moving_img[425:435, 300:320] = 8
+
+    tx = sitkutils.fft_based_translation_initialization(
+        fixed_img,
+        moving_img,
+    )
+    assert tx.GetOffset() == (425, 300.0)
+
+    tx = sitkutils.fft_based_translation_initialization(
+        fixed_img, moving_img, required_fraction_of_overlapping_pixels=0.5
+    )
+    assert tx.GetOffset() == (-85.0, 50.0)
+
+
+def test_fft_initialization4():
+    """Testing with initial transform."""
+    fixed_img = sitk.Image([1024, 512], sitk.sitkUInt8)
+    fixed_img[510:520, 255:265] = 10
+
+    moving_img = sitk.Image([1024, 512], sitk.sitkUInt8)
+    moving_img.SetSpacing((10, 10))
+    moving_img[425:435, 300:310] = 8
+
+    initial_transform = sitk.Similarity2DTransform(10)
+
+    tx = sitkutils.fft_based_translation_initialization(
+        fixed_img, moving_img, initial_transform=initial_transform
+    )
+    assert tx.GetTranslation() == (-850.0, 450.0)
+
+
 def test_overlay_bounding_boxes():
     bounding_boxes = [[10, 10, 60, 20], [200, 180, 230, 250]]
     scalar_image = sitk.Image([256, 256], sitk.sitkUInt8)