Example Detect Describe Interface

From BoofCV
Revision as of 13:58, 17 January 2022 by Peter (talk | contribs)
(diff) ← Older revision | Latest revision (diff) | Newer revision → (diff)
Jump to navigationJump to search

BoofCV provides multiple ways to detect and describe interest points inside of images. The easiest high level interface to work with is DetectDescribePoint. It will detect and describe all interest points in the image at the same time. The alternative involves using separate interfaces for detection, orientation, and describing.

Example Code:


  • Interest point detection
  • Local region descriptors

Related Examples

Example Code

 * {@link DetectDescribePoint} provides a single unified interface for detecting interest points inside of images
 * and describing the features. For some features (e.g. SIFT) it can be much faster than the alternative approach
 * where individual algorithms are used for feature detection, orientation estimation, and describe. It also
 * simplifies the code.
 * This example demonstrates how to create instances, but the {@link ExampleAssociatePoints} demonstrates how
 * to use the interface.
 * @author Peter Abeles
public class ExampleDetectDescribe {

	 * For some features, there are pre-made implementations of DetectDescribePoint. This has only been done
	 * in situations where there was a performance advantage or that it was a very common combination.
	public static <T extends ImageGray<T>, TD extends TupleDesc<TD>>
	DetectDescribePoint<T, TD> createFromPremade( Class<T> imageType ) {
		var config = new ConfigFastHessian();
		config.maxFeaturesPerScale = 200;
		return (DetectDescribePoint)FactoryDetectDescribe.surfStable(config, null, null, imageType);
//		var config = new ConfigCompleteSift();
//		config.detector.maxFeaturesPerScale = 400;
//		return (DetectDescribePoint)FactoryDetectDescribe.sift(config, imageType);

	 * Any arbitrary implementation of InterestPointDetector, OrientationImage, DescribeRegionPoint
	 * can be combined into DetectDescribePoint. The syntax is more complex, but the end result is more flexible.
	 * This should only be done if there isn't a pre-made DetectDescribePoint.
	public static <T extends ImageGray<T>, TD extends TupleDesc<TD>>
	DetectDescribePoint<T, TD> createFromComponents( Class<T> imageType ) {
		// create a corner detector
		Class derivType = GImageDerivativeOps.getDerivativeType(imageType);
		GeneralFeatureDetector corner = FactoryDetectPoint.createShiTomasi(new ConfigGeneralDetector(1000, 5, 1), null, derivType);
		InterestPointDetector detector = FactoryInterestPoint.wrapPoint(corner, 1, imageType, derivType);

		// describe points using BRIEF
		DescribePointRadiusAngle describe = FactoryDescribePointRadiusAngle.brief(new ConfigBrief(true), imageType);

		// Combine together.
		// NOTE: orientation will not be estimated
		return FactoryDetectDescribe.fuseTogether(detector, null, describe);

	public static void main( String[] args ) {

		Class imageType = GrayF32.class;

		DetectDescribePoint detDesc = createFromPremade(imageType);
//		DetectDescribePoint detDesc = createFromComponents(imageType);

		// Might as well have this example do something useful, like associate two images
		ScoreAssociation scorer = FactoryAssociation.defaultScore(detDesc.getDescriptionType());
		AssociateDescription associate = FactoryAssociation.greedy(new ConfigAssociateGreedy(true), scorer);

		// load and match images
		var app = new ExampleAssociatePoints(detDesc, associate, imageType);

		BufferedImage imageA = UtilImageIO.loadImageNotNull(UtilIO.pathExample("stitch/kayak_01.jpg"));
		BufferedImage imageB = UtilImageIO.loadImageNotNull(UtilIO.pathExample("stitch/kayak_03.jpg"));

		app.associate(imageA, imageB);