# This file is part of Sympathy for Data.
# Copyright (c) 2017, Combine Control Systems AB
#
# SYMPATHY FOR DATA COMMERCIAL LICENSE
# You should have received a link to the License with Sympathy for Data.
import numpy as np
from packaging import version
from sympathy.api import node
from sympathy.api.nodeconfig import Port, Ports, Tag, Tags
from sympathy.api.exceptions import (
SyNodeError, SyDataError, SyConfigurationError)
import scipy
from scipy.spatial.distance import cdist
from skimage import feature
from sylib.imageprocessing.image import ImagePort
scipy_version = version.Version(scipy.__version__)
[docs]
class MatchTemplate(node.Node):
"""
Uses cross-correlation to match a template image to an image.
Returns an image with maxima where the template matches.
"""
name = 'Match Template'
author = 'Mathias Broxvall'
icon = 'image_template.svg'
description = (
'Uses cross-correlation to match a template image to an image. '
'Returns an image with maxima where the template matches.'
)
nodeid = 'com.sympathyfordata.imageanalysis.match_template'
tags = Tags(Tag.ImageProcessing.ImageManipulation)
parameters = node.parameters()
parameters.set_string(
'padding mode', value='No padding',
description=(
'Method used for padding the input'),
label='Padding mode',
editor=node.editors.combo_editor(
options=['No padding', 'constant', 'edge', 'reflect', 'wrap'])
)
parameters.set_float(
'k', value=0.0,
description=(
'Value used for padding mode "constant"'),
label='Pad value'
)
inputs = Ports([
ImagePort('image to perform matching on', name='image'),
ImagePort('Template image', name='template'),
])
outputs = Ports([
ImagePort('result after matching', name='result'),
])
def execute(self, node_context):
image = node_context.input['image'].get_image()
template = node_context.input['template'].get_image()
pad_mode = node_context.parameters['padding mode'].value
pad_value = node_context.parameters['k'].value
result = feature.match_template(
image, template,
pad_input=True if pad_mode != 'No padding' else False,
mode=pad_mode if pad_mode != 'No padding' else 'constant',
constant_values=pad_value)
node_context.output['result'].set_image(result)
[docs]
class MatchDescriptors(node.Node):
"""
Performs brute-force matching between keypoints in the two tables
"""
author = 'Mathias Broxvall'
icon = 'image_match_descriptors.svg'
description = ('Performs brute-force matching between keypoints in '
'the two tables')
name = 'Match Descriptors'
tags = Tags(Tag.ImageProcessing.Extract)
nodeid = 'com.sympathyfordata.imageanalysis.match_descriptors'
parameters = node.parameters()
parameters.set_string(
'rescolumns', value='X, Y', label='Result columns',
description=(
'Names of columns to be ignored during matching, separated '
'by a comma. These columns will instead be included in the '
'final output'),
)
parameters.set_string(
'prefix1', value='1:', label='Prefix 1',
description=('Prefix added to output names for columns from '
'first input. '),
)
parameters.set_string(
'prefix2', value='2:', label='Prefix 2',
description=('Prefix added to output names for columns from '
'second input. '),
)
_minkowski = 'minkowski'
parameters.set_string(
'metric', value='', label='Metric',
description=(
'Metric used for comparing keypoints. '
'See scipy.spatial.distance.cdist for details.'),
editor=node.editors.combo_editor(
options=['euclidean', 'cityblock', 'correlation', _minkowski,
'hamming', '', 'braycurtis', 'canberra',
'chebyshev', 'cosine', 'dice', 'hamming', 'jaccard',
'kulsinski', 'mahalanobis', 'matching', 'rogerstanimoto',
'russellrao', 'seuclidean', 'sokalmichener',
'sokalsneath', 'sqeuclidean', 'yule'])
)
parameters.set_boolean(
'cross', value=False, label='Cross-check',
description=(
'Performs matching in both direction and return a match '
'(key1, key2) only if keypoint 2 is the best match for keypoint 1 '
'and keypoint 1 is the best match for keypoint 2'),
)
parameters.set_float(
'max_distance', value=np.inf, label='Max distance',
description='Maximum allowed distance to be regarded as a match')
parameters.set_integer(
'p', value=2, label='P',
description='P norm to apply for minkowski metrics')
controllers = node.controller(
when=node.field('metric', 'value', value=_minkowski),
action=(node.field('p', 'enabled')))
inputs = Ports([
Port.Table('Keypoints 1', name='keypoints 1'),
Port.Table('Keypoints 2', name='keypoints 2'),
])
outputs = Ports([
Port.Table('Table with results', name='result'),
])
def execute(self, node_context):
input1 = node_context.input['keypoints 1']
input2 = node_context.input['keypoints 2']
output = node_context.output['result']
prefix1 = node_context.parameters['prefix1'].value
prefix2 = node_context.parameters['prefix2'].value
pnorm = node_context.parameters['p'].value
cross_check = node_context.parameters['cross'].value
metric = node_context.parameters['metric'].value
max_distance = node_context.parameters['max_distance'].value
if input1.number_of_rows() == 0 or input2.number_of_rows() == 0:
raise SyDataError("Empty table")
rescol = node_context.parameters['rescolumns'].value
rescol = [s.strip() for s in rescol.split(',')]
res1 = [col.data for col in input1.cols() if col.name in rescol]
res2 = [col.data for col in input2.cols() if col.name in rescol]
res1_names = [prefix1 + col.name
for col in input1.cols() if col.name in rescol]
res2_names = [prefix2 + col.name
for col in input2.cols() if col.name in rescol]
desc1 = np.column_stack(
[col.data for col in input1.cols() if col.name not in rescol])
desc2 = np.column_stack(
[col.data for col in input2.cols() if col.name not in rescol])
if desc1.shape[1] != desc2.shape[1]:
raise SyNodeError("Number of input descriptors does not match")
if metric == '':
if np.issubdtype(desc1.dtype, np.bool_):
metric = 'hamming'
else:
metric = 'euclidean'
elif metric == 'wminkowski':
raise SyConfigurationError(
f'Metric: {metric} is no longer supported, use '
f'{self._minkowski} instead?')
if metric == 'kulsinski' and scipy_version >= version.Version('1.9.0'):
metric = 'kulczynski1'
if metric == self._minkowski:
distances = cdist(desc1, desc2, metric=metric, p=pnorm)
else:
distances = cdist(desc1, desc2, metric=metric)
indices1 = np.arange(desc1.shape[0])
indices2 = np.argmin(distances, axis=1)
if cross_check:
matches1 = np.argmin(distances, axis=0)
mask = indices1 == matches1[indices2]
indices1 = indices1[mask]
indices2 = indices2[mask]
if max_distance < np.inf:
mask = distances[indices1, indices2] < max_distance
indices1 = indices1[mask]
indices2 = indices2[mask]
match_dist = distances[indices1, indices2]
output.set_column_from_array("match", match_dist)
for name, col in zip(res1_names, res1):
output.set_column_from_array(name, col[indices1])
for name, col in zip(res2_names, res2):
output.set_column_from_array(name, col[indices2])
