Difference between revisions of "Example Remove Perspective Distortion"

From BoofCV
Jump to navigationJump to search
m
m
Line 9: Line 9:


Example Code:
Example Code:
* [https://github.com/lessthanoptimal/BoofCV/blob/v0.19/examples/src/boofcv/examples/geometry/ExampleRemovePerspectiveDistortion.java ExampleRemovePerspectiveDistortion.java]
* [https://github.com/lessthanoptimal/BoofCV/blob/v0.20/examples/src/boofcv/examples/geometry/ExampleRemovePerspectiveDistortion.java ExampleRemovePerspectiveDistortion.java]


Concepts:
Concepts:
Line 24: Line 24:
/**
/**
  * Certain image processing techniques, such as Optical Character Recognition (OCR), can be performed better if
  * Certain image processing techniques, such as Optical Character Recognition (OCR), can be performed better if
  * perspective distortion is remove from an image.  In this example a homography is computed from the cour corners
  * perspective distortion is remove from an image.  In this example a homography is computed from the four corners
  * of a bulletin board and the image is projected into a square image without perspective distortion.
  * of a bulletin board and the image is projected into a square image without perspective distortion. The
* {@link RemovePerspectiveDistortion} class is used to perform the distortion.  The class is easy to understand
* if you know what a homography is, you should look at it!
  *
  *
  * @author Peter Abeles
  * @author Peter Abeles
Line 33: Line 35:


// load a color image
// load a color image
BufferedImage buffered = UtilImageIO.loadImage("../data/applet/goals_and_stuff.jpg");
BufferedImage buffered = UtilImageIO.loadImage(UtilIO.pathExample("goals_and_stuff.jpg"));
MultiSpectral<ImageFloat32> input = ConvertBufferedImage.convertFromMulti(buffered, null, true, ImageFloat32.class);
MultiSpectral<ImageFloat32> input = ConvertBufferedImage.convertFromMulti(buffered, null, true, ImageFloat32.class);


// Create a smaller output image for processing later on
RemovePerspectiveDistortion<MultiSpectral<ImageFloat32>> removePerspective =
MultiSpectral<ImageFloat32> output = input._createNew(400,500);
new RemovePerspectiveDistortion<MultiSpectral<ImageFloat32>>(400,500, ImageType.ms(3,ImageFloat32.class));


// Homography estimation algorithm. Requires a minimum of 4 points
// Specify the corners in the input image of the region.
Estimate1ofEpipolar computeHomography = FactoryMultiView.computeHomography(true);
// Order matters! top-left, top-right, bottom-right, bottom-left
if( !removePerspective.apply(input,
new Point2D_F64(267, 182), new Point2D_F64(542, 68),
new Point2D_F64(519, 736), new Point2D_F64(276, 570)) ){
throw new RuntimeException("Failed!?!?");
}


// Specify the pixel coordinates from destination to target
MultiSpectral<ImageFloat32> output = removePerspective.getOutput();
ArrayList<AssociatedPair> associatedPairs = new ArrayList<AssociatedPair>();
associatedPairs.add(new AssociatedPair(new Point2D_F64(0,0),new Point2D_F64(267,182)));
associatedPairs.add(new AssociatedPair(new Point2D_F64(output.width-1,0),new Point2D_F64(542,68)));
associatedPairs.add(new AssociatedPair(new Point2D_F64(output.width-1,output.height-1),new Point2D_F64(519,736)));
associatedPairs.add(new AssociatedPair(new Point2D_F64(0,output.height-1),new Point2D_F64(276,570)));
 
// Compute the homography
DenseMatrix64F H = new DenseMatrix64F(3,3);
computeHomography.process(associatedPairs, H);
 
// Create the transform for distorting the image
PointTransformHomography_F32 homography = new PointTransformHomography_F32(H);
PixelTransform_F32 pixelTransform = new PointToPixelTransform_F32(homography);
 
// Apply distortion and show the results
DistortImageOps.distortMS(input,output,pixelTransform, null, TypeInterpolate.BILINEAR);


BufferedImage flat = ConvertBufferedImage.convertTo_F32(output,null,true);
BufferedImage flat = ConvertBufferedImage.convertTo_F32(output,null,true);

Revision as of 09:41, 9 November 2015

It is often easier to process an image after perspective distortion is removed. The billboard in this example is at an acute angle relative to the camera, making its text hard to read. A homography can be created, using the for cornerns of the billboard, and used to remove this distortion.

Example Code:

Concepts:

  • Homography
  • Perpsective
  • Distortion

Relevant Examples:

Example Code

/**
 * Certain image processing techniques, such as Optical Character Recognition (OCR), can be performed better if
 * perspective distortion is remove from an image.  In this example a homography is computed from the four corners
 * of a bulletin board and the image is projected into a square image without perspective distortion.  The
 * {@link RemovePerspectiveDistortion} class is used to perform the distortion.  The class is easy to understand
 * if you know what a homography is, you should look at it!
 *
 * @author Peter Abeles
 */
public class ExampleRemovePerspectiveDistortion {
	public static void main(String[] args) {

		// load a color image
		BufferedImage buffered = UtilImageIO.loadImage(UtilIO.pathExample("goals_and_stuff.jpg"));
		MultiSpectral<ImageFloat32> input = ConvertBufferedImage.convertFromMulti(buffered, null, true, ImageFloat32.class);

		RemovePerspectiveDistortion<MultiSpectral<ImageFloat32>> removePerspective =
				new RemovePerspectiveDistortion<MultiSpectral<ImageFloat32>>(400,500, ImageType.ms(3,ImageFloat32.class));

		// Specify the corners in the input image of the region.
		// Order matters! top-left, top-right, bottom-right, bottom-left
		if( !removePerspective.apply(input,
				new Point2D_F64(267, 182), new Point2D_F64(542, 68),
				new Point2D_F64(519, 736), new Point2D_F64(276, 570)) ){
			throw new RuntimeException("Failed!?!?");
		}

		MultiSpectral<ImageFloat32> output = removePerspective.getOutput();

		BufferedImage flat = ConvertBufferedImage.convertTo_F32(output,null,true);
		ShowImages.showWindow(buffered,"Original Image",true);
		ShowImages.showWindow(flat,"Without Perspective Distortion",true);
	}
}