def threshold_otsu_median(datafield, pts=256):
"""
Compute Otsu's threshold with median classifier (histogram mode).
Parameters
----------
datafield: gwy.DataField
Datafield over which threshold will be computed.
pts: int
Number of points in histogram. Default 256.
Returns
-------
threshold: float
"""
# number of pts for histogram
cdf = gwy.DataLine(pts, 0, True)
# compute z-scale CDF from image
datafield.cdh(cdf, pts)
# sort all intensity values
vals = sorted(datafield.get_data())
# sort out hist data
dx = cdf.get_real() / cdf.get_res() # pixel spacing in data line
offset = cdf.get_offset() # dataline offset, ie. 0 index point value in x
x = [(i * dx) + offset for i in range(pts)] # construct bin locations
# get cdf data as list
weight1 = [int(round(i * len(vals))) for i in cdf.get_data()]
weight2 = [len(vals) - i for i in weight1[::-1]][::-1]
# class medians for all possible thresholds
median1 = [vals[:weight1[i]][weight1[i] // 2] for i in range(pts - 1)]
median2 = [
vals[::-1][:weight2[i]][weight2[i] // 2] for i in range(pts - 1)
]
# median for whole image
medianT = vals[len(vals) // 2]
# Clip ends to align class 1 and class 2 variables:
# The last value of `weight1`/`mean1` should pair with zero values in
# `weight2`/`mean2`, which do not exist.
variance = [
weight1[i] * (median1[i] - medianT)**2 + weight2[i + 1] *
(median2[i] - medianT)**2 for i in range(pts - 1)
]
# get index of maximum variance to index x array
threshold = x[:-1][variance.index(max(variance))]
return threshold
def threshold_otsu_mode(datafield, pts=256):
"""
Compute Otsu's threshold with median classifier (histogram mode).
Parameters
----------
datafield: gwy.DataField
Datafield over which threshold will be computed.
pts: int
Number of points in histogram. Default 256.
Returns
-------
threshold: float
"""
# number of pts for histogram
dl = gwy.DataLine(pts, 0, True)
# compute z-scale histogram from image
datafield.dh(dl, pts)
# sort out hist data
dx = dl.get_real() / dl.get_res() # pixel spacing in data line
offset = dl.get_offset() # dataline offset, ie. 0 index point value in x
x = [(i * dx) + offset for i in range(pts)] # construct bin locations
y = dl.get_data() # get histogram values
weights = dl.duplicate()
weights.cumulate() # calculate cdf
weights.multiply(dx) # make sure cdf sums to 1
weights = weights.get_data() # get cdf data as list
# holder lists for modal values at different thresholds
mode1 = []
mode2 = []
modeT = x[y.index(dl.get_max())] # get modal value for distribution
# for each possible threshold value in histogram
for i in range(1, pts):
mode1.append(x[y.index(dl.part_extract(0, i).get_max())])
mode2.append(x[i + y[i:].index(dl.part_extract(i, pts - i).get_max())])
# calculate intraclass variance depending on the thresholds used
variance = []
for i in range(len(mode1)): # should be same as mode2, ie. pts-1
variance.append(weights[:-1][i] * (mode1[i] - modeT)**2 +
(1 - weights[1:][i]) * (mode2[i] - modeT)**2)
# get index of maximum variance to index x array
threshold = x[:-1][variance.index(max(variance))]
return threshold
def line_correct_median(data):
""" corrected = line_correct_median(data)
Does a median line correction on the passed gwy.DataField.
"""
xres = data.get_xres()
yres = data.get_yres()
line = gwy.DataLine(xres,1.0,False)
modi = gwy.DataLine(yres,1.0,False)
for i in range(yres):
data.get_row(line,i)
median = line.get_median()
modi.set_val(i,median)
median = modi.get_median()
for i in range(yres):
data.area_add(0, i, xres, 1, median - modi.get_val(i))
return data