# -*- coding: utf-8 -*-
"""
IlluminationCorrection
======================
The module provides a function to correct systematic illumination variations
and vignetting in intensity.
The intensity image :math:`I(x)` given a flat field :math:`F(x)` and
a background :math:`B(x)` the image is corrected to :math:`C(x)` as:
.. math:
C(x) = \\frac{I(x) - B(x)}{F(x) - B(x)}
The module also has functionality to create flat field corections from measured
intensity changes in a single direction, useful e.g. for lightsheet images,
see e.g. :func:`flatfield_from_regression`.
References
----------
..[LSM] Fundamentals of Light Microscopy and Electronic Imaging, p. 421
"""
__author__ = 'Christoph Kirst <christoph.kirst.ck@gmail.com>'
__license__ = 'GPLv3 - GNU General Pulic License v3 (see LICENSE.txt)'
__copyright__ = 'Copyright © 2020 by Christoph Kirst'
__webpage__ = 'http://idisco.info'
__download__ = 'http://www.github.com/ChristophKirst/ClearMap2'
import numpy as np
import os
from scipy.optimize import curve_fit
import matplotlib.pyplot as plt
import ClearMap.IO.IO as io
import ClearMap.Utils.Timer as tmr
import ClearMap.Utils.HierarchicalDict as hdict
import ClearMap.Settings as settings
###############################################################################
### Default parameter
###############################################################################
default_flat_field_line_file_name = os.path.join(settings.resources_path, "Microscope/lightsheet_flatfield_correction.csv");
"""Default file of points along the illumination changing line for the flat field correction
See Also:
:func:`correct_illumination`
"""
###############################################################################
### Illuminaton correction
###############################################################################
[docs]def correct_illumination(source, flatfield = None, background = None, scaling = None, dtype = None, verbose = False):
"""Correct illumination and background.
Arguments
---------
source : array, str, or Source
The image to correct for illumination.
sink : array, str, Source, or None
The sink to write results to.
flatfield : str, array, Source or None
The flatfield estimate. If None, no flat field correction is done.
background : str, array, Source or None
The background estimate. If None, backgorund is assumed to be zero.
scaling : float, 'max', 'mean' or None
Scale the corrected result by this factor. If 'max' or 'mean' scale the
result to match the 'max' or 'mean'. If None, dont scale the result.
processes : int or None
Number of processes to use. If None use macimal available.
verbose : bool
If true, print progrss infomration.
Returns
-------
corrected : array
Illumination corrected image.
Note
----
The intensity image :math:`I(x)` given a flat field :math:`F(x)` and
a background :math:`B(x)` image is corrected to :math:`C(x)` as:
.. math:
C(x) = \\frac{I(x) - B(x)}{F(x) - B(x)}
If the background is not given :math:`B(x) = 0`.
The correction is done slice by slice assuming the data was collected with
a light sheet microscope.
The image is finally optionally scaled.
References
----------
[1] Fundamentals of Light Microscopy and Electronic Imaging, p 421
See Also
--------
:const:`default_flatfield_line_file_name`
"""
if background is not None:
background = io.as_source(background);
if flatfield is None:
return source;
if flatfield is True:
# default flatfield correction
flatfield = default_flat_field_line_file_name;
if isinstance(flatfield, str):
flatfield = io.as_source(flatfield);
if flatfield.ndim == 1:
flatfield = flatfield_from_line(flatfield, source.shape[1]);
if flatfield.shape[:2] != source.shape[:2]:
raise ValueError("The flatfield shape %r does not match the source shape %r!" % (flatfield.shape[:2], source.shape[:2]));
flatfield = io.as_source(flatfield);
if verbose:
timer = tmr.Timer();
hdict.pprint(head = 'Illumination correction:', flatfield=flatfield, background=background, scaling=scaling);
#initilaize source
source = io.as_source(source);
if dtype is None:
dtype = source.dtype;
# rescale factor
flatfield = flatfield.array.astype(dtype);
if scaling is True:
scaling = "mean";
if isinstance(scaling, str):
if scaling.lower() == "mean":
# scale back by average flat field correction:
scaling = flatfield.mean();
elif scaling.lower() == "max":
scaling = flatfield.max();
else:
raise RuntimeError('Scaling not "max" or "mean" but %r!' % (scaling,));
# illumination correction in each slice
corrected = np.array(source.array, dtype=dtype)
if background is None:
for z in range(source.shape[2]):
corrected[:,:,z] = source[:,:,z] / flatfield;
else:
if background.shape != flatfield.shape:
raise RuntimeError("Illumination correction: background does not match source shape: %r vs %r!" % (background.shape, source[:,:,0].shape));
background = background.array.astype('float32');
flatfield = (flatfield - background);
for z in range(source.shape[2]):
corrected[:,:,z] = (source[:,:,z] - background) / flatfield;
if not scaling is None:
corrected= corrected * scaling;
if verbose:
timer.print_elapsed_time('Illumination correction');
return corrected
[docs]def flatfield_from_line(line, shape, axis = 0, dtype = float):
"""Creates a 2d flat field image from a 1d line of estimated intensities.
Arguments
---------
line : array
Array of intensities along the specified axis.
shape : tuple
Shape of the resulting image.
axis : int
Axis of the flat field line estimate.
Returns
-------
flatfield : array
Full 2d flat field.
"""
line = io.as_source(line);
if isinstance(shape, int):
shape = (line.shape[0], shape) if axis == 0 else (shape, line.shape[0]);
if shape[axis] != line.shape[0]:
raise ValueError('Line shape %d does not match image shape %d!' % (line.shape[0], shape[axis]));
shape = shape[axis];
flatfield = np.array([line.array] * shape, dtype=dtype)
if axis == 1:
flatfield = flatfield.T;
return flatfield;
[docs]def flatfield_line_from_regression(source, sink = None, positions = None, method = 'polynomial', reverse = None, return_function = False, verbose = False):
"""Create flat field line fit from a list of positions and intensities.
Arguments
---------
source : str, array or Source
Intensities as (n,)-vector or (n,m)-array of m intensity measurements
at n points along an axis.
sink : str, array, Source or None
Sink for the result.
positions : array, 'source' or None
The positions of the soource points. If None, a linear increasing
positions with equal spaccing is assumed. If 'source' take positions from
first line of the source array.
method : 'Gaussian' or 'Polynomial'
function type for the fit.
reverse : bool
Reverse the line fit after fitting.
return_function : bool
If True, also return the fitted function.
verbose :bool
Print and plot information for the fit.
Returns
-------
fit : array
Fitted intensities on points.
fit_function : function
Fitted function.
Note
----
The fit is either to be assumed to be a 'Gaussian':
.. math:
I(x) = a \\exp^{- (x- x_0)^2 / (2 \\sigma)) + b"
or follows a order 6 radial 'Polynomial'
.. math:
I(x) = a + b (x- x_0)^2 + c (x- x_0)^4 + d (x- x_0)^6
"""
source = io.as_source(source);
# split source
if source.ndim == 1:
y = np.atleast_2d(source.array);
elif source.ndim == 2:
if positions == 'source':
positions = source[:,0];
y = source[:,1:-1];
else:
y = source.array;
else:
raise RuntimeError('flatfield_line_from_regression: input data not a line or array of x,i data');
if positions is None:
positions = np.arange(source.shape[0])
#calculate mean of the intensity measurements
x = positions;
ym = np.mean(y, axis = 1);
if verbose > 1:
plt.figure()
for i in range(1,source.shape[1]):
plt.plot(x, source[:,i]);
plt.plot(x, ym, 'k');
if method.lower() == 'polynomial':
## fit r^6
mean = sum(ym * x)/sum(ym)
def f(x,m,a,b,c,d):
return a + b * (x-m)**2 + c * (x-m)**4 + d * (x-m)**6;
popt, pcov = curve_fit(f, x, ym, p0 = (mean, 1, 1, 1, .1));
m = popt[0]; a = popt[1]; b = popt[2];
c = popt[3]; d = popt[4];
def fopt(x):
return f(x, m = m, a = a, b = b, c = c, d = d);
if verbose:
print("polynomial fit: %f + %f (x- %f)^2 + %f (x- %f)^4 + %f (x- %f)^6" % (a, b, m, c, m, d, m));
else:
## Gaussian fit
mean = sum(ym * x)/sum(ym)
sigma = sum(ym * (x-mean)**2)/(sum(ym))
def f(x, a, m, s, b):
return a * np.exp(- (x - m)**2 / 2 / s) + b;
popt, pcov = curve_fit(f, x, ym, p0 = (1000, mean, sigma, 400));
a = popt[0]; m = popt[1]; s = popt[2]; b = popt[3];
def fopt(x):
return f(x, a=a, m=m, s=s, b=b);
if verbose:
print("Gaussian fit: %f exp(- (x- %f)^2 / (2 %f)) + %f" % (a, m, s, b));
fit = fopt(x);
if reverse:
fit.reverse();
if verbose > 1:
plt.plot(x, fit);
plt.title('flatfield_line_from_regression')
result = io.write(sink, fit);
if return_function:
result = (result, fopt)
return result;
###############################################################################
### Tests
###############################################################################
def _test():
"""Tests"""
import ClearMap.Visualization.Plot3d as p3d
import ClearMap.ImageProcessing.IlluminationCorrection as ic
ff = ic.flatfield_from_line(ic.default_flat_field_line_file_name, 100);
p3d.plot(ff)
Source
