Scripting the Trainable Weka Segmentation

Revision as of 13:53, 17 April 2013 by Iarganda (talk | contribs) (added basic script commands: load image into WekaSegmentation and add labels)

Scripting is one of the reasons Fiji is so powerful, and the Trainable Segmentation library (that includes the Advanced Weka Segmentation plugin methods) is one of the best examples for scriptable Fiji components.

Getting started

The first thing you need to start scripting the is to know which methods you can use. For that, please have a look at the API of the Trainable Segmentation library, which is available here.

Let's go through the basic commands with examples written in Beanshell:

In order to include all the library methods, the easiest (but not elegant) way of doing it is importing the whole library:

import trainableSegmentation.*;

Now we are ready to play. We can open our input image and assign it to a WekaSegmentation object or segmentator:

// input train image
input  = IJ.openImage( "input-grayscale-or-color-image.tif" );
// create Weka Segmentation object
segmentator = new WekaSegmentation( input );

At it is now, the segmentator has default parameters and default classifier. That means that it will use the same features that are set by default in the Advanced Weka Segmentation plugin, 2 classes (named "class 1" and "class 2") and a random forest classifier with 200 trees and 2 random features per node. If we are fine with that, we can now add some labels for our training data and train the classifier based on them.

There are different ways of adding labels to our data:

1) we can add any type of ROI to any of the existing classes using "addExample":

// add pixels to first class (0) from ROI in slice # 1
segmentator.addExample( 0, new Roi( 10, 10, 50, 50 ), 1 );
// add pixels to second class (1) from ROI in slice # 1
segmentator.addExample( 1, new Roi( 400, 400, 30, 30 ), 1 );

2) add the labels from a binary image, where white pixels belong to one class and black pixels belong to the other class:

// open binary label image
labels  = IJ.openImage( "binary-labels.tif" );
// for slice #1, add white pixels as labels for class 2 and black pixels as labels for class 1
segmentator.addBinaryData( labels, 1, "class 2", "class 1");

Define your own features

Here is a little Javascript that makes two features from the Clown example and uses them to train a classifier (see the inline comments for more information):


var image = IJ.openImage(System.getProperty("ij.dir") + "/samples/clown.jpg");
if (image.getStackSize() > 1)
        new StackConverter(image).convertToGray32();

var duplicator = new Duplicator();

// process the image into different stacks, one per feature:

var smoothed =;, "Gaussian Blur...", "radius=20");

var medianed =;, "Median...", "radius=10");

// add new feature here (1/2)

// the FeatureStackArray contains a FeatureStack for every slice in our original image
var featuresArray = new FeatureStackArray(image.getStackSize(), 1, 16, false,
        1, 19, null);

// turn the list of stacks into FeatureStack instances, one per original
// slice. Each FeatureStack contains exactly one slice per feature.
for (var slice = 1; slice <= image.getStackSize(); slice++) {
        var stack = new ImageStack(image.getWidth(), image.getHeight());
        stack.addSlice("smoothed", smoothed.getStack().getProcessor(slice));
        stack.addSlice("medianed", medianed.getStack().getProcessor(slice));

        // add new feature here (2/2) and do not forget to add it with a
        // unique slice label!

        var featuresImage = new ImagePlus("slice " + slice, stack);
        var features = new FeatureStack(featuresImage);
        featuresArray.set(features, slice - 1);

var wekaSegmentation = new WekaSegmentation(image);

// set examples for class 1 (= foreground) and 0 (= background))
function addExample(classNum, slice, xArray, yArray) {
        var polygon = new FloatPolygon(xArray, yArray);
        var roi = new PolygonRoi(polygon, PolygonRoi.FREELINE);
        IJ.log("roi: " + roi);
        wekaSegmentation.addExample(classNum, roi, slice);

 * generate these with the macro:

        getSelectionCoordinates(x, y);

        print('['); Array.print(x); print('],");
        print('['); Array.print(y); print(']");
addExample(1, 1,
        [ 82,85,85,86,87,87,87,88,88,88,88,88,88,88,88,86,86,84,83,82,81,
          54,53,51,50,49,49,49,51,52,53,54,55,55,56,56 ],
        [ 141,137,136,134,133,132,130,129,128,127,126,125,124,123,122,121,
          121,123,125,126,128,128,129,129,130 ]
addExample(0, 1,
        [ 167,165,163,161,158,157,157,157,157,157,157,157,158 ],
        [ 30,29,29,29,29,29,28,26,25,24,23,22,21 ]

// train classifier
if (!wekaSegmentation.trainClassifier())
        throw new RuntimeException("Uh oh! No training today.");

output = wekaSegmentation.applyClassifier(image);;

Define binary labels programmatically

Here is a simple script in Beanshell doing the following:

  1. It takes one image (2D or stack) as training input image and a binary image as the corresponding labels.
  2. Train a classifier (in this case a random forest, but it can be changed) based on randomly selected pixels of the training image. The number of samples (pixels to use for training) is also a parameter, and it will be the same for each class.
  3. Apply the trained classifier to a test image (2D or stack).
import ij.*;
import ij.process.*;
import trainableSegmentation.*;
import hr.irb.fastRandomForest.*;

// training input image (it could be a stack or a single 2D image)
image = IJ.openImage("train-volume.tif");
// corresponding binary labels
labels = IJ.openImage("train-labels.tif");

// create Weka segmentator
seg = new WekaSegmentation(image);

// number of samples to use per slice
nSamplesToUse = 2000;

// Classifier
// In this case we use a Fast Random Forest
rf = new FastRandomForest();
// Number of trees in the forest
// Number of features per tree
// Seed
rf.setSeed( (new java.util.Random()).nextInt() );

// set classifier

// Parameters 
// membrane patch size
// maximum radius of the filters

// Selected attributes
enableFeatures = new boolean[]{
			true, 	/* Gaussian_blur */
			true, 	/* Sobel_filter */
			true, 	/* Hessian */
			true, 	/* Difference_of_gaussians */
			true, 	/* Membrane_projections */
			false, 	/* Variance */
			false, 	/* Mean */
			false, 	/* Minimum */
			false, 	/* Maximum */
			false, 	/* Median */
			false,	/* Anisotropic_diffusion */
			false, 	/* Bilateral */
			false, 	/* Lipschitz */
			false, 	/* Kuwahara */
			false,	/* Gabor */
			false, 	/* Derivatives */
			false, 	/* Laplacian */
			false,	/* Structure */
			false,	/* Entropy */
			false	/* Neighbors */

// Enable features in the segmentator
seg.setEnabledFeatures( enableFeatures );

// Add balanced labels
seg.addRandomBalancedBinaryData(image, labels, "class 2", "class 1", nSamplesToUse);

// Train classifier

// Un-comment next 7 lines to segment and display training image
// Apply classifier to current image
seg.applyClassifier( true );

// Display classified image
prob = seg.getClassifiedImage();
prob.setTitle( "Probability maps of train image" );;

// Open test image
image = IJ.openImage("test-volume.tif");

// Apply trained classifier to test image and get probabilities
prob = seg.applyClassifier(image, 0, true );

// Display results
prob.setTitle( "Probability maps of test image" );;;