I adapted Volko Straub's script to use Jython scripting because it gave me more control and extended an example from the Jython scripting page (link in code). The main point is that one can access the results table. One could make this more complicated by only drawing the well separated particles that are not touching boundaries on the original image. I do this frequently in my particle sizing work. There are also ways to tune the watershed a bit better. I should warn you that Python/Jython is sensitive to indentation. As you can see from the initial part, one needs to import the functions one needs. This helps avoid namespace clashes. Anyway, I just did a small filter based upon circularity and wrote out the equivalent circular diameter values (in pixels) to a .csv.
I typically loop over a series of images, and dump out a feature vector for each blob of interest into a .csv file. I did some very simple processing here. I usually start with a very lenient classifier, dump the data to a .csv and take a look at the data using R. This permits better classification and automated report generation. i have done this with both R/RStudio and lately some with iPython notebooks.
Since all these Open Source tools are generally text based, they all (mostly) play well with version control and permit frequently-needed functions to be accumulated in packages to make our work more reproducible. Best wishes as you go down this path.
John Minter
Eastman Kodak
Analytical Sciences Microscopy Lab
# procTouchingLatex.py
import os
import math
from ij import IJ, ImagePlus
from ij.process import ImageProcessor
from ij.process import AutoThresholder
from ij.process.AutoThresholder import getThreshold
from ij.process.AutoThresholder.Method import Otsu
from ij.measure import ResultsTable
from ij.plugin.filter import EDM
# I use environment variables and store images and reports is predictable directories.
# To test, you can just make a path to a working directory. Note that Windows, Mac
# and Linux will let you use a forward slash...
homDir = os.environ['HOME']
relPrj = "/dat/images/IJ/"
wrkDir = homDir + relPrj
# put main config parameters at the top....
minCirc = 0.840 # see num 5 below
iDigits = 4 # for rounding
sampID = 'latex-TEM'
bVerbose = False
sImgPath = wrkDir + sampID + '.gif'
sCsvPath = wrkDir + sampID + ".csv"
ori = ImagePlus(sImgPath)
ori.show()
# Make a copy of the original image, doing the conversion to 8 bit:
# adapted from IJ Jython Scripting page
http://imagej.net/Jython_Scriptingwrk = ori.createImagePlus()
ip = ori.getProcessor().duplicate().convertToByteProcessor()
wrk.setProcessor("Latex", ip)
stats = wrk.getStatistics()
his = stats.histogram
# 2 - Apply a threshold: only zeros and ones
#
# This is from Otsu, set manually
thr = AutoThresholder().getThreshold(Otsu, his)
if bVerbose:
print(thr)
ip.setThreshold(0, thr, ImageProcessor.NO_LUT_UPDATE)
# Call the Thresholder to convert the image to a mask
IJ.run(wrk, "Convert to Mask", "")
# 3 - Apply watershed
# Create and run new EDM object, which is an Euclidean Distance Map (EDM)
# and run the watershed on the ImageProcessor:
EDM().toWatershed(ip)
# 4 - Show the watershed image
wrk.show()
# 5 set up the measurements and compute them. Writes them to the results
# table. Compare particles 1, 75,76, and 77 (partially extracted spheres) to the others.
# "Good" spheres have a circularity of > about 0.83
IJ.run(wrk, "Set Measurements...", "area centroid center fit shape redirect=None decimal=3")
IJ.run(wrk, "Analyze Particles...", "display exclude clear add")
# 6 - Get the results table and compute the ECD for the "Good" Particles
rt = ResultsTable().getResultsTable()
lArea = rt.getColumn(rt.getColumnIndex("Area"))
lCirc = rt.getColumn(rt.getColumnIndex("Circ."))
nPart = len(lArea)
partID = []
partECD = []
k=0
for i in range(nPart):
area = lArea[i]
circ = lCirc[i]
ecd = 2.0 * math.sqrt(area/math.pi)
if circ > minCirc:
partID.append(i+1)
partECD.append(round(ecd, iDigits))
if bVerbose:
print(i+1, ecd)
# write the output file as .csv
f=open(sCsvPath, 'w')
strLine = 'part, ecd.px\n'
f.write(strLine)
for k in range(len(partECD)):
strLine = "%d, %.5f\n" % (partID[k], partECD[k] )
f.write(strLine)
f.close()
Locked
