Example Thresholding

• Variable Lighting
• 

  Calibration grid. Left: original, Middle: Global Otsu, Right: Local Square

• Difficult Text Example
• 

  Left: Original, Middle: Local Square, Right: Sauvola

Thresholding gray scale images is one of the most basic ways to segment an image. It is quick and effective in many situations. BoofCV provides several algorithms for computing both global and locally adaptive thresholds.

Example Code:

• ExampleThresholding.java

Concepts:

• Segmentation
• Thresholding

Relevant Examples/Tutorials:

• Tutorial Image Segmentation
• Example Binary Image
• Example Superpixels
• Example Color Segmentation

Example Code

/**
 * Demonstration of different techniques for automatic thresholding an image to create a binary image.  The binary
 * image can then be used for shape analysis and other applications.  Global methods apply the same threshold
 * to the entire image.  Local methods compute a local threshold around each pixel and can handle uneven
 * lighting, but produce noisy results in regions with uniform lighting.
 *
 * @see boofcv.examples.imageprocessing.ExampleBinaryOps
 *
 * @author Peter Abeles
 */
public class ExampleThresholding {

	public static void threshold( String imageName ) {
		BufferedImage image = UtilImageIO.loadImage(imageName);

		// convert into a usable format
		GrayF32 input = ConvertBufferedImage.convertFromSingle(image, null, GrayF32.class);
		GrayU8 binary = new GrayU8(input.width,input.height);

		// Display multiple images in the same window
		ListDisplayPanel gui = new ListDisplayPanel();

		// Global Methods
		GThresholdImageOps.threshold(input, binary, ImageStatistics.mean(input), true);
		gui.addImage(VisualizeBinaryData.renderBinary(binary, false, null),"Global: Mean");
		GThresholdImageOps.threshold(input, binary, GThresholdImageOps.computeOtsu(input, 0, 255), true);
		gui.addImage(VisualizeBinaryData.renderBinary(binary, false, null),"Global: Otsu");
		GThresholdImageOps.threshold(input, binary, GThresholdImageOps.computeEntropy(input, 0, 255), true);
		gui.addImage(VisualizeBinaryData.renderBinary(binary, false, null),"Global: Entropy");

		// Local method
		GThresholdImageOps.localSquare(input, binary, 28, 1.0, true, null, null);
		gui.addImage(VisualizeBinaryData.renderBinary(binary, false, null),"Local: Square");
		GThresholdImageOps.localBlockMinMax(input, binary, 10, 1.0, true, 15 );
		gui.addImage(VisualizeBinaryData.renderBinary(binary, false, null),"Local: Block Min-Max");
		GThresholdImageOps.localGaussian(input, binary, 42, 1.0, true, null, null);
		gui.addImage(VisualizeBinaryData.renderBinary(binary, false, null),"Local: Gaussian");
		GThresholdImageOps.localSauvola(input, binary, 5, 0.30f, true);
		gui.addImage(VisualizeBinaryData.renderBinary(binary, false, null),"Local: Sauvola");

		// Sauvola is tuned for text image.  Change radius to make it run better in others.

		// Show the image image for reference
		gui.addImage(ConvertBufferedImage.convertTo(input,null),"Input Image");

		String fileName =  imageName.substring(imageName.lastIndexOf('/')+1);
		ShowImages.showWindow(gui,fileName);
	}

	public static void main(String[] args) {
		// example in which global thresholding works best
		threshold(UtilIO.pathExample("particles01.jpg"));
		// example in which adaptive/local thresholding works best
		threshold(UtilIO.pathExample("segment/uneven_lighting_squares.jpg"));
		// hand written text with non-uniform stained background
		threshold(UtilIO.pathExample("segment/stained_handwriting.jpg"));
	}
}