Applying a saved classifier returns a black image when running script

[J-na]

Hello, i want to use Trainable Weka Segmentation to analyse images of fluorescent cell clusters.
When manually running the plugin i can load a saved classifier and apply it just fine.
However, when automating the segmentation using a beanshell script it returns a black image instead as the result. This is the script used:

import ij.IJ;
import ij.ImageStack;
import ij.ImagePlus;
import ij.process.ColorSpaceConverter;
import ij.process.ByteProcessor;
import trainableSegmentation.*;
import weka.clusterers.EM;
import weka.clusterers.SimpleKMeans;
import weka.core.WekaPackageManager;
import weka.core.WekaPackageClassLoaderManager;
// input train image
input = IJ.getImage();
// create Weka Segmentation object
segmentator = new WekaSegmentation( input );
// load classifier from file
segmentator.loadClassifier("D:/fiji-win64/greenClass.model");
// apply classifier to current training image and get label result
// (set parameter to true to get probabilities)
segmentator.applyClassifier(false);
// get result (float image)
result = segmentator.getClassifiedImage();
result.show();

The script does return the probability maps as expected , but not the classified image . The used image and expected results are shown here:

The original image

The result from manual segmentation

The result from running the script with segmentator.applyClassifier(false); , this happens for every image processed using the script

The probability from running the script with segmentator.applyClassifier(true);

I am not sure if this is an issue with the script i am using or with the trainable segmentation itself, but i am hoping it is an easy fix. Thanks in advance :)

Edit: I also tried to immediately convert the generated probability using this script posted earlier https://gist.github.com/iarganda/c7fc0a88b8d2737c9d3d , which generated the same black image as generated by segmentator.applyClassifier(false);. Perhaps it is something related to this?

[iarganda]

Hello @J-na,

Are you sure the image is completely "black"? It might be just 0s and 1s so you need to adjust the contrast or simply apply a LUT to see both classes. If you want to have the same LUT as in the plugin, after getting the result image, do the following:

import trainableSegmentation.utils.Utils;

result.setLut( Utils.getGoldenAngleLUT() ); 

You have more information here.

[J-na]

Hello @iarganda,

Thank you for getting back to me so quick.
Intensity analysis confirms that the image contains only pixels with 0 intensity. Applying the golden angle LUT results in this image.

A quick test of the 'color-based segmentation using clustering' seems to produce promising results so i will look into using that instead.