def align(self, dlg):
''' Executes the alignment. If the alignment is successful, the aligned stream is
added to the main window. If not, an error message is shown.
dlg (AlignmentAcquisitionDialog): The plugin dialog
'''
crop = (self.crop_top.value, self.crop_bottom.value,
self.crop_left.value, self.crop_right.value)
flip = (self.flip_x.value, self.flip_y.value)
tem_img = preprocess(self._nem_proj.raw[0], self.invert.value, flip,
crop, self.blur.value, True)
sem_raw = img.ensure2DImage(self._rem_proj.raw[0])
sem_img = preprocess(sem_raw, False, (False, False), (0, 0, 0, 0),
self.blur_ref.value, True)
try:
tmat, _, _, _, _ = keypoint.FindTransform(tem_img, sem_img)
# get the metadata corresponding to the transformation
transf_md = get_img_transformation_md(tmat, tem_img, sem_img)
logging.debug("Computed transformation metadata: %s", transf_md)
except ValueError as ex:
box = wx.MessageDialog(dlg, str(ex), "Failed to align images",
wx.OK | wx.ICON_STOP)
box.ShowModal()
box.Destroy()
return
# Shear is really big => something is gone wrong
if abs(transf_md[model.MD_SHEAR]) > 1:
logging.warning(
"Shear is %g, which means the alignment is probably wrong",
transf_md[model.MD_SHEAR])
transf_md[model.MD_SHEAR] = 0
# Pixel size ratio is more than 2 ? => something is gone wrong
# TODO: pixel size 100x bigger/smaller than the reference is also wrong
pxs = transf_md[model.MD_PIXEL_SIZE]
if not (0.5 <= pxs[0] / pxs[1] <= 2):
logging.warning(
"Pixel size is %s, which means the alignment is probably wrong",
pxs)
transf_md[model.MD_PIXEL_SIZE] = (pxs[0], pxs[0])
# The actual image inserted is not inverted and not blurred, but we still
# want it flipped and cropped.
raw = preprocess(self._nem_proj.raw[0], False, flip, crop, 0, False)
raw.metadata.update(transf_md)
# Add a new stream panel (removable)
analysis_tab = self.main_app.main_data.getTabByName('analysis')
aligned_stream = stream.StaticSEMStream(
self._nem_proj.stream.name.value, raw)
scont = analysis_tab.stream_bar_controller.addStream(aligned_stream,
add_to_view=True)
scont.stream_panel.show_remove_btn(True)
# Finish by closing the window
dlg.Close()
def test_no_match(self):
'''Testing the matching of two feature-less images'''
image = numpy.zeros((1000, 1000), dtype=numpy.uint8) + 86
timg = numpy.zeros((2000, 1000), dtype=numpy.uint16) + 2563
image = preprocess(image, False, (False, False), (0, 0, 0, 0), 0, True)
timg = preprocess(timg, False, (False, False), (0, 0, 0, 0), 0, True)
with self.assertRaises(ValueError):
tmat, _, _, _, _ = keypoint.FindTransform(timg, image)
def test_no_match(self):
'''Testing the matching of two feature-less images'''
image = numpy.zeros((1000, 1000), dtype=numpy.uint8) + 86
timg = numpy.zeros((2000, 1000), dtype=numpy.uint16) + 2563
image = preprocess(image, False, (False, False), (0, 0, 0, 0), 0, True)
timg = preprocess(timg, False, (False, False), (0, 0, 0, 0), 0, True)
with self.assertRaises(ValueError):
tmat, _, _, _, _ = keypoint.FindTransform(timg, image)
def align(self, dlg):
''' Executes the alignment. If the alignment is successful, the aligned stream is
added to the main window. If not, an error message is shown.
dlg (AlignmentAcquisitionDialog): The plugin dialog
'''
crop = (self.crop_top.value, self.crop_bottom.value,
self.crop_left.value, self.crop_right.value)
flip = (self.flip_x.value, self.flip_y.value)
tem_img = preprocess(self._nem_proj.raw[0], self.invert.value, flip, crop,
self.blur.value, True)
sem_raw = img.ensure2DImage(self._rem_proj.raw[0])
sem_img = preprocess(sem_raw, False, (False, False), (0, 0, 0, 0),
self.blur_ref.value, True)
try:
tmat, _, _, _, _ = keypoint.FindTransform(tem_img, sem_img)
# get the metadata corresponding to the transformation
transf_md = get_img_transformation_md(tmat, tem_img, sem_img)
logging.debug("Computed transformation metadata: %s", transf_md)
except ValueError as ex:
box = wx.MessageDialog(dlg, str(ex), "Failed to align images",
wx.OK | wx.ICON_STOP)
box.ShowModal()
box.Destroy()
return
# Shear is really big => something is gone wrong
if abs(transf_md[model.MD_SHEAR]) > 1:
logging.warning("Shear is %g, which means the alignment is probably wrong",
transf_md[model.MD_SHEAR])
transf_md[model.MD_SHEAR] = 0
# Pixel size ratio is more than 2 ? => something is gone wrong
# TODO: pixel size 100x bigger/smaller than the reference is also wrong
pxs = transf_md[model.MD_PIXEL_SIZE]
if not (0.5 <= pxs[0] / pxs[1] <= 2):
logging.warning("Pixel size is %s, which means the alignment is probably wrong",
pxs)
transf_md[model.MD_PIXEL_SIZE] = (pxs[0], pxs[0])
# The actual image inserted is not inverted and not blurred, but we still
# want it flipped and cropped.
raw = preprocess(self._nem_proj.raw[0], False, flip, crop, 0, False)
raw.metadata.update(transf_md)
# Add a new stream panel (removable)
analysis_tab = self.main_app.main_data.getTabByName('analysis')
aligned_stream = stream.StaticSEMStream(self._nem_proj.stream.name.value, raw)
scont = analysis_tab.stream_bar_controller.addStream(aligned_stream, add_to_view=True)
scont.stream_panel.show_remove_btn(True)
# Finish by closing the window
dlg.Destroy()
def _precompute_kp(self):
if self.draw_kp.value:
if not self._nem_proj or not self._rem_proj:
return
# # TODO: pass extra args for the keypoint detector
# dtkargs = {"WTA_K": self.wta.value,
# "scaleFactor": self.scaleFactor.value,
# "nlevels": self.nlevels.value,
# "patchSize": self.patchSize.value,
# "edgeThreshold": self.patchSize.value, # should be equal
# }
crop = (self.crop_top.value, self.crop_bottom.value,
self.crop_left.value, self.crop_right.value)
flip = (self.flip_x.value, self.flip_y.value)
tem_img = preprocess(self._nem_proj.raw[0], self.invert.value,
flip, crop, self.blur.value, True)
sem_raw = img.ensure2DImage(self._rem_proj.raw[0])
sem_img = preprocess(sem_raw, False, (False, False), (0, 0, 0, 0),
self.blur_ref.value, True)
try:
tmat, self._nem_kp, self._rem_kp, self._nem_mkp, self._rem_mkp = \
keypoint.FindTransform(tem_img, sem_img)
except ValueError as ex:
logging.debug("No match found: %s", ex)
# TODO: if no match, still show the keypoints
self._nem_kp = None
self._nem_mkp = None
self._rem_kp = None
self._rem_mkp = None
else:
self._nem_kp = None
self._nem_mkp = None
self._rem_kp = None
self._rem_mkp = None
self._update_ref_stream()
self._update_new_stream()
def _precompute_kp(self):
if self.draw_kp.value:
if not self._nem_proj or not self._rem_proj:
return
# # TODO: pass extra args for the keypoint detector
# dtkargs = {"WTA_K": self.wta.value,
# "scaleFactor": self.scaleFactor.value,
# "nlevels": self.nlevels.value,
# "patchSize": self.patchSize.value,
# "edgeThreshold": self.patchSize.value, # should be equal
# }
crop = (self.crop_top.value, self.crop_bottom.value,
self.crop_left.value, self.crop_right.value)
flip = (self.flip_x.value, self.flip_y.value)
tem_img = preprocess(self._nem_proj.raw[0], self.invert.value, flip, crop,
self.blur.value, True)
sem_raw = img.ensure2DImage(self._rem_proj.raw[0])
sem_img = preprocess(sem_raw, False, (False, False), (0, 0, 0, 0),
self.blur_ref.value, True)
try:
tmat, self._nem_kp, self._rem_kp, self._nem_mkp, self._rem_mkp = \
keypoint.FindTransform(tem_img, sem_img)
except ValueError as ex:
logging.debug("No match found: %s", ex)
# TODO: if no match, still show the keypoints
self._nem_kp = None
self._nem_mkp = None
self._rem_kp = None
self._rem_mkp = None
else:
self._nem_kp = None
self._nem_mkp = None
self._rem_kp = None
self._rem_mkp = None
self._update_ref_stream()
self._update_new_stream()
def _updateImage(self):
raw = self.stream.raw[0]
metadata = self.stream._find_metadata(raw.metadata)
raw = img.ensure2DImage(raw) # Remove extra dimensions (of length 1)
grayscale_im = preprocess(raw, self._invert, self._flip, self._crop,
self._gaussian_sigma, self._eqhis)
rgb_im = img.DataArray2RGB(grayscale_im)
if self._kp:
rgb_im = cv2.drawKeypoints(rgb_im, self._kp, None, color=(30, 30, 255), flags=0)
if self._mkp:
rgb_im = cv2.drawKeypoints(rgb_im, self._mkp, None, color=(0, 255, 0), flags=0)
rgb_im = model.DataArray(rgb_im, metadata)
rgb_im.flags.writeable = False
self.image.value = rgb_im
def test_image_pair(self):
''' Testing a pair of images
'''
# WARNING: if opencv is not compiled with SIFT support (ie, only ORB
# available), then this test case will fail.
# FIXME: these two images are very hard, and any tiny change in the
# algorithm or settings can cause the alignment to fail => not a good
# test case
# only one image will be used, but this structure helps to test
# different images
image_pairs = [
(
('Slice69_stretched.tif', True, (False, True), (0, 0, 0, 0), 6),
('g_009_cropped.tif', False, (False, False), (0, 0, 0, 0), 3)
),
# (
# ('001_CBS_010.tif', False, (False, False), (0, 0, 0, 0), 0),
# ('20141014-113042_1.tif', False, (False, False), (0, 0, 0, 0), 0)
# ),
# (
# ('t3 DELPHI.tiff', False, (False, False), (0, 200, 0, 0), 3),
# ('t3 testoutA3.tif', False, (False, False), (0, 420, 0, 0), 3)
# )
]
image_pair = image_pairs[0]
# open the images
tem_img = open_acquisition(os.path.join(IMG_PATH, image_pair[0][0]))[0].getData()
sem_img = open_acquisition(os.path.join(IMG_PATH, image_pair[1][0]))[0].getData()
# preprocess
tem_img = preprocess(tem_img, image_pair[0][1], image_pair[0][2],
image_pair[0][3], image_pair[0][4], True)
sem_img = preprocess(sem_img, image_pair[1][1], image_pair[1][2],
image_pair[1][3], image_pair[1][4], True)
# execute the algorithm to find the transform between the images
tmat, _, _, _, _ = keypoint.FindTransform(tem_img, sem_img)
# uncomment this if you want to see the keypoint images
'''tem_painted_kp = cv2.drawKeypoints(tem_img, tem_kp, None, color=(0,255,0), flags=0)
sem_painted_kp = cv2.drawKeypoints(sem_img, sem_kp, None, color=(0,255,0), flags=0)
cv2.imwrite(IMG_PATH + 'tem_kp.jpg', tem_painted_kp)
cv2.imwrite(IMG_PATH + 'sem_kp.jpg', sem_painted_kp)'''
# uncomment this if you want to see the warped image
'''warped_im = cv2.warpPerspective(tem_img, tmat, (sem_img.shape[1], sem_img.shape[0]))
merged_im = cv2.addWeighted(sem_img, 0.5, warped_im, 0.5, 0.0)
cv2.imwrite(IMG_PATH + 'merged_with_warped.jpg', merged_im)'''
tmetadata = get_img_transformation_md(tmat, tem_img, sem_img)
logging.debug("Computed metadata = %s", tmetadata)
# FIXME: these values are actually pretty bad
# comparing based on a successful alignment validated from the warped image
# self.assertAlmostEqual(8.7e-07, tmetadata[model.MD_PIXEL_SIZE][0], places=6)
# self.assertAlmostEqual(1.25e-06, tmetadata[model.MD_PIXEL_SIZE][1], places=6)
# self.assertAlmostEqual(0.085, tmetadata[model.MD_ROTATION], places=2)
# self.assertAlmostEqual(0.000166, tmetadata[model.MD_POS][0], places=5)
# self.assertAlmostEqual(-0.0001435, tmetadata[model.MD_POS][1], places=5)
# self.assertAlmostEqual(0.035, tmetadata[model.MD_SHEAR], places=2)
#
# Check that calling the function again with the same data returns the
# same results (bug happens when using FLANN-KDtree matcher)
for i in range(2):
tmatn, _, _, _, _ = keypoint.FindTransform(tem_img, sem_img)
tmetadatan = get_img_transformation_md(tmatn, tem_img, sem_img)
logging.debug("Computed metadata = %s", tmetadatan)
numpy.testing.assert_equal(tmatn, tmat)
self.assertEqual(tmetadatan, tmetadata)
def test_synthetic_images(self):
''' Testing the matching of a synthetic image. The image is generated with
a rotation and scale, and then it checks if the matching algorithm
came up with the same result
'''
# generate a syntyetic image
image = numpy.zeros((1000, 1000, 4), dtype=numpy.uint8)
surface = cairo.ImageSurface.create_for_data(image, cairo.FORMAT_ARGB32, 1000, 1000)
cr = cairo.Context(surface)
cr.set_source_rgb(1.0, 1.0, 1.0)
cr.paint()
cr.set_source_rgb(0.0, 0.0, 0.0)
# draw circles
cr.arc(200, 150, 80, 0, 2 * math.pi)
cr.fill()
cr.arc(400, 150, 70, 0, 2 * math.pi)
cr.fill()
cr.arc(700, 180, 50, 0, 2 * math.pi)
cr.fill()
cr.arc(200, 500, 80, 0, 2 * math.pi)
cr.fill()
cr.arc(400, 600, 70, 0, 2 * math.pi)
cr.fill()
cr.arc(600, 500, 50, 0, 2 * math.pi)
cr.fill()
cr.arc(600, 500, 50, 0, 2 * math.pi)
cr.fill()
cr.arc(500, 500, 350, 0, 2 * math.pi)
cr.set_line_width(5)
cr.stroke()
cr.arc(600, 500, 50, 0, 2 * math.pi)
cr.fill()
# center circle
cr.arc(500, 500, 5, 0, 2 * math.pi)
cr.fill()
# rectangle
cr.rectangle(600, 700, 200, 100)
cr.fill()
image = image[:, :, 0]
angle = 0.3
scale = 0.7
translation_x = 100
translation_y = 50
# generate a rotation/scale matrix, with the rotation centered on the center of the image
rot_scale_mat = cv2.getRotationMatrix2D((500.0, 500.0), math.degrees(angle), scale)
# generate the transformed image with scale and rotation
timg = cv2.warpAffine(image, rot_scale_mat, (1000, 1000),
borderMode=cv2.BORDER_CONSTANT, borderValue=(255, 255, 255))
# generate a transformation matrix with translation
translation_mat = numpy.float32([[1, 0, translation_x], [0, 1, translation_y]])
# generate the transformed image with translation
timg = cv2.warpAffine(timg, translation_mat, (1000, 1000),
borderMode=cv2.BORDER_CONSTANT, borderValue=(255, 255, 255))
image = preprocess(image, False, (False, False), (0, 0, 0, 0), 0, True)
timg = preprocess(timg, False, (False, False), (0, 0, 0, 0), 0, True)
# execute the matching algorithm, and find the transformation matrix between the original
# and the transformed image
tmat_odemis, _, _, _, _ = keypoint.FindTransform(timg, image)
timg_md = {}
timg = model.DataArray(timg, timg_md)
image_md = {
model.MD_PIXEL_SIZE: (1e-6, 1e-6),
model.MD_POS: (35e-6, 25e-6),
model.MD_ROTATION: 0.15,
model.MD_SHEAR: 0.15
}
image = model.DataArray(image, image_md)
# use the invert matrix to get the original values
tmetadata = get_img_transformation_md(inv(tmat_odemis), timg, image)
logging.debug("Computed metadata = %s", tmetadata)
# the matching algorithm is not that accurate, so the values are approximated
self.assertAlmostEqual(0.7e-6, tmetadata[model.MD_PIXEL_SIZE][0], places=7)
self.assertAlmostEqual(0.7e-6, tmetadata[model.MD_PIXEL_SIZE][1], places=7)
# 0.3 + 0.15
self.assertAlmostEqual(0.45, tmetadata[model.MD_ROTATION], places=1)
# (100 + 35) * PS
self.assertAlmostEqual(135e-06, tmetadata[model.MD_POS][0], places=5)
# (-50 + 25) * PS
self.assertAlmostEqual(-25e-06, tmetadata[model.MD_POS][1], places=5)
# 0.0 (there's no shear on the image) + 0.15
self.assertAlmostEqual(0.15, tmetadata[model.MD_SHEAR], places=1)
# uncomment this if you want to see the images used on this test
'''
def test_image_pair(self):
''' Testing a pair of images
'''
# WARNING: if opencv is not compiled with SIFT support (ie, only ORB
# available), then this test case will fail.
# FIXME: these two images are very hard, and any tiny change in the
# algorithm or settings can cause the alignment to fail => not a good
# test case
# only one image will be used, but this structure helps to test
# different images
image_pairs = [
(
('Slice69_stretched.tif', True, (False, True), (0, 0, 0, 0), 6),
('g_009_cropped.tif', False, (False, False), (0, 0, 0, 0), 3)
),
# (
# ('001_CBS_010.tif', False, (False, False), (0, 0, 0, 0), 0),
# ('20141014-113042_1.tif', False, (False, False), (0, 0, 0, 0), 0)
# ),
# (
# ('t3 DELPHI.tiff', False, (False, False), (0, 200, 0, 0), 3),
# ('t3 testoutA3.tif', False, (False, False), (0, 420, 0, 0), 3)
# )
]
image_pair = image_pairs[0]
# open the images
tem_img = open_acquisition(os.path.join(IMG_PATH, image_pair[0][0]))[0].getData()
sem_img = open_acquisition(os.path.join(IMG_PATH, image_pair[1][0]))[0].getData()
# preprocess
tem_img = preprocess(tem_img, image_pair[0][1], image_pair[0][2],
image_pair[0][3], image_pair[0][4], True)
sem_img = preprocess(sem_img, image_pair[1][1], image_pair[1][2],
image_pair[1][3], image_pair[1][4], True)
# execute the algorithm to find the transform between the images
try:
tmat, _, _, _, _ = keypoint.FindTransform(tem_img, sem_img)
except ValueError:
if not hasattr(cv2, 'SIFT') and not hasattr(cv2, 'SIFT_create'):
self.skipTest("Test only works with SIFT, not with ORB.")
else:
raise AssertionError("Failed to find transform between images.")
# uncomment this if you want to see the keypoint images
'''tem_painted_kp = cv2.drawKeypoints(tem_img, tem_kp, None, color=(0,255,0), flags=0)
sem_painted_kp = cv2.drawKeypoints(sem_img, sem_kp, None, color=(0,255,0), flags=0)
cv2.imwrite(IMG_PATH + 'tem_kp.jpg', tem_painted_kp)
cv2.imwrite(IMG_PATH + 'sem_kp.jpg', sem_painted_kp)'''
# uncomment this if you want to see the warped image
'''warped_im = cv2.warpPerspective(tem_img, tmat, (sem_img.shape[1], sem_img.shape[0]))
merged_im = cv2.addWeighted(sem_img, 0.5, warped_im, 0.5, 0.0)
cv2.imwrite(IMG_PATH + 'merged_with_warped.jpg', merged_im)'''
tmetadata = get_img_transformation_md(tmat, tem_img, sem_img)
logging.debug("Computed metadata = %s", tmetadata)
# FIXME: these values are actually pretty bad
# comparing based on a successful alignment validated from the warped image
# self.assertAlmostEqual(8.7e-07, tmetadata[model.MD_PIXEL_SIZE][0], places=6)
# self.assertAlmostEqual(1.25e-06, tmetadata[model.MD_PIXEL_SIZE][1], places=6)
# self.assertAlmostEqual(0.085, tmetadata[model.MD_ROTATION], places=2)
# self.assertAlmostEqual(0.000166, tmetadata[model.MD_POS][0], places=5)
# self.assertAlmostEqual(-0.0001435, tmetadata[model.MD_POS][1], places=5)
# self.assertAlmostEqual(0.035, tmetadata[model.MD_SHEAR], places=2)
#
# Check that calling the function again with the same data returns the
# same results (bug happens when using FLANN-KDtree matcher)
for i in range(2):
try:
tmatn, _, _, _, _ = keypoint.FindTransform(tem_img, sem_img)
except ValueError:
if not hasattr(cv2, 'SIFT') and not hasattr(cv2, 'SIFT_create'):
self.skipTest("Test only works with SIFT, not with ORB.")
else:
raise AssertionError("Failed to find transform between images.")
tmetadatan = get_img_transformation_md(tmatn, tem_img, sem_img)
logging.debug("Computed metadata = %s", tmetadatan)
numpy.testing.assert_equal(tmatn, tmat)
self.assertEqual(tmetadatan, tmetadata)
def test_synthetic_images(self):
''' Testing the matching of a synthetic image. The image is generated with
a rotation and scale, and then it checks if the matching algorithm
came up with the same result
'''
# generate a syntyetic image
image = numpy.zeros((1000, 1000, 4), dtype=numpy.uint8)
surface = cairo.ImageSurface.create_for_data(image, cairo.FORMAT_ARGB32, 1000, 1000)
cr = cairo.Context(surface)
cr.set_source_rgb(1.0, 1.0, 1.0)
cr.paint()
cr.set_source_rgb(0.0, 0.0, 0.0)
# draw circles
cr.arc(200, 150, 80, 0, 2 * math.pi)
cr.fill()
cr.arc(400, 150, 70, 0, 2 * math.pi)
cr.fill()
cr.arc(700, 180, 50, 0, 2 * math.pi)
cr.fill()
cr.arc(200, 500, 80, 0, 2 * math.pi)
cr.fill()
cr.arc(400, 600, 70, 0, 2 * math.pi)
cr.fill()
cr.arc(600, 500, 50, 0, 2 * math.pi)
cr.fill()
cr.arc(600, 500, 50, 0, 2 * math.pi)
cr.fill()
cr.arc(500, 500, 350, 0, 2 * math.pi)
cr.set_line_width(5)
cr.stroke()
cr.arc(600, 500, 50, 0, 2 * math.pi)
cr.fill()
# center circle
cr.arc(500, 500, 5, 0, 2 * math.pi)
cr.fill()
# rectangle
cr.rectangle(600, 700, 200, 100)
cr.fill()
image = image[:, :, 0]
angle = 0.3
scale = 0.7
translation_x = 100
translation_y = 50
# generate a rotation/scale matrix, with the rotation centered on the center of the image
rot_scale_mat = cv2.getRotationMatrix2D((500.0, 500.0), math.degrees(angle), scale)
# generate the transformed image with scale and rotation
timg = cv2.warpAffine(image, rot_scale_mat, (1000, 1000),
borderMode=cv2.BORDER_CONSTANT, borderValue=(255, 255, 255))
# generate a transformation matrix with translation
translation_mat = numpy.float32([[1, 0, translation_x], [0, 1, translation_y]])
# generate the transformed image with translation
timg = cv2.warpAffine(timg, translation_mat, (1000, 1000),
borderMode=cv2.BORDER_CONSTANT, borderValue=(255, 255, 255))
image = preprocess(image, False, (False, False), (0, 0, 0, 0), 0, True)
timg = preprocess(timg, False, (False, False), (0, 0, 0, 0), 0, True)
# execute the matching algorithm, and find the transformation matrix between the original
# and the transformed image
tmat_odemis, _, _, _, _ = keypoint.FindTransform(timg, image)
timg_md = {}
timg = model.DataArray(timg, timg_md)
image_md = {
model.MD_PIXEL_SIZE: (1e-6, 1e-6),
model.MD_POS: (35e-6, 25e-6),
model.MD_ROTATION: 0.15,
model.MD_SHEAR: 0.15
}
image = model.DataArray(image, image_md)
# use the invert matrix to get the original values
tmetadata = get_img_transformation_md(inv(tmat_odemis), timg, image)
logging.debug("Computed metadata = %s", tmetadata)
# the matching algorithm is not that accurate, so the values are approximated
self.assertAlmostEqual(0.7e-6, tmetadata[model.MD_PIXEL_SIZE][0], places=7)
self.assertAlmostEqual(0.7e-6, tmetadata[model.MD_PIXEL_SIZE][1], places=7)
# 0.3 + 0.15
self.assertAlmostEqual(0.45, tmetadata[model.MD_ROTATION], places=1)
# (100 + 35) * PS
self.assertAlmostEqual(135e-06, tmetadata[model.MD_POS][0], places=5)
# (-50 + 25) * PS
self.assertAlmostEqual(-25e-06, tmetadata[model.MD_POS][1], places=5)
# 0.0 (there's no shear on the image) + 0.15
self.assertAlmostEqual(0.15, tmetadata[model.MD_SHEAR], places=1)
# uncomment this if you want to see the images used on this test
'''