TIEVisionFeaturesDetector

Declaration

TIEVisionFeaturesDetector = interface(TIEVisionBase)

Description

An interface for keypoint detection and matching between two images, using ORB, SIFT, AKAZE, KAZE or BRISK (see TIEVisionFeatureAlgorithm).
The matching points can be used to align one image to another, e.g. with findHomography and warpPerspective.

Create it with createFeaturesDetector.

Which algorithm works best depends on the images, so try another if alignment fails or is inaccurate. In tests aligning the three image pairs of the ImageAlignment demo with maxFeatures = 2000:
◼All five algorithms aligned all three pairs accurately
◼ORB took about 60 ms, SIFT 300 ms, BRISK 400 ms, AKAZE 550 ms and KAZE 1400 ms
With maxFeatures = 500, ORB was less accurate on one pair and BRISK failed on one pair.

Note: Large images are reduced before detection to limit memory use (except with ORB). See setMaxImageSize.


Note:
◼A shortcut method for this is available: DetectAndDeskew
◼The legacy class TIEVisionLibrary.createORBFeaturesDetector returns the same interface using ORB (TIEVisionORBFeaturesDetector is an alias of TIEVisionFeaturesDetector)

Demos

Examples

Also see: Automated Samples

var
  bmp1, bmp2: TIEBitmap;
  detector: TIEVisionFeaturesDetector;
  hmat: TIEVisionMatrix;
  imgres: TIEVisionImage;
begin
  bmp1 := TIEBitmap.Create('scanned-form.jpg');
  bmp2 := TIEBitmap.Create('form.jpg');

  // detect matching points of the two images
  detector := IEVisionLib().createFeaturesDetector( ievFeatureSIFT );
  detector.detect( bmp1.GetIEVisionImage(), bmp2.GetIEVisionImage(), 2000 );

  // optionally show matching graph (matching points links)
  //ImageEnView2.IEBitmap.AssignIEVisionImage( detector.getMatchingGraph() );

  // find homography matrix (hmat) from matching points
  hmat := IEVisionLib().createMath().findHomography( detector.getPoints1(), detector.getPoints2(), ievRANSAC );

  // adjust "scanned-form" to the same skew of "form" image
  imgres := bmp1.GetIEVisionImage().warpPerspective( hmat, IEVisionSize( bmp2.Width, bmp2.Height ));

  // show final result
  ImageEnView1.IEBitmap.AssignIEVisionImage( imgres );

  bmp2.Free();
  bmp1.Free();
end;


// Find matching keypoints between the image and a rotated (15 deg.) and scaled (80%) copy
detector := IEVisionLib().createFeaturesDetector( ievFeatureSIFT );
detector.detect( ImageEnView1.IEBitmap.GetIEVisionImage(), ImageEnView2.IEBitmap.GetIEVisionImage(), 2000 );
ImageEnView1.IEBitmap.AssignIEVisionImage( detector.getMatchingGraph() );
ImageEnView1.Update(); // Show changes in our viewer

Methods and Properties

See Also

◼createFeaturesDetector
◼findHomography
◼isValidHomography
◼warpPerspective

---