boofcv.alg.shapes.polygon

Class BinaryPolygonConvexDetector<T extends boofcv.struct.image.ImageSingleBand>

java.lang.Object

boofcv.alg.shapes.polygon.BinaryPolygonConvexDetector<T>

public class BinaryPolygonConvexDetector<T extends boofcv.struct.image.ImageSingleBand>
extends java.lang.Object

Detects convex polygons with the specified number of sides in an image. Shapes are assumed to be black shapes against a white background, allowing for thresholding to be used.

Processing Steps:

1. First the input a gray scale image and a binarized version of it.
2. The contours of black blobs are found.
3. From the contours polygons are fitted and refined to pixel accuracy.
4. (Optional) Sub-pixel refinement of the polygon's edges and/or corners.

Subpixel refinement is done using the provided RefinePolygonLineToImage and RefinePolygonCornersToImage. If straight lines are straight, as is the case with images with lens distortion removes, then the line based refinement will tend to produce better results. If lens distortion is present then the corner sub-pixel algorithm is more likely to produce better results.

NOTE: If both refinement methods are used then corner is applied first followed by line.
NOTE: A binary image is not processed as input because the gray-scale image is used in the sub-pixel line/corner refinement.

The returned polygons will encompass the entire black polygon. Here is a simple example in 1D. If all pixels are white, but pixels ranging from 5 to 10, inclusive, then the returned boundaries would be 5.0 to 11.0. This means that coordinates 5.0 ≤ x < 11.0 are all black. 11.0 is included, but note that the entire pixel 11 is white.

Field Summary

Fields

Modifier and Type	Field and Description
protected boofcv.struct.distort.PixelTransform_F32	toDistorted
protected boofcv.struct.distort.PixelTransform_F32	toUndistorted

Constructor Summary

Constructors

Constructor and Description
BinaryPolygonConvexDetector(int[] polygonSides, SplitMergeLineFitLoop contourToPolygon, PolygonEdgeScore differenceScore, RefinePolygonLineToImage<T> refineLine, RefinePolygonCornersToImage<T> refineCorner, double minContourFraction, double splitDistanceFraction, boolean outputClockwise, java.lang.Class<T> inputType) Configures the detector.

Method Summary

Modifier and Type	Method and Description
org.ddogleg.struct.FastQueue<georegression.struct.shapes.Polygon2D_F64>	getFound()
java.util.List<boofcv.alg.filter.binary.Contour>	getFoundContours()
java.lang.Class<T>	getInputType()
boofcv.struct.image.ImageSInt32	getLabeled()
int[]	getNumberOfSides()
boolean	isCheckEdgeBefore()
boolean	isOutputClockwise()
void	process(T gray, boofcv.struct.image.ImageUInt8 binary) Examines the undistorted gray scake input image for squares.
void	setCheckEdgeBefore(boolean checkEdgeBefore) If set to true it will prune using polygons using their edge intensity before sub-pixel optimization.
void	setLensDistortion(int width, int height, boofcv.struct.distort.PixelTransform_F32 toUndistorted, boofcv.struct.distort.PixelTransform_F32 toDistorted) Specifies transforms which can be used to change coordinates from distorted to undistorted and the opposite coordinates.
void	setNumberOfSides(int[] numberOfSides)
void	setVerbose(boolean verbose)
protected boolean	touchesBorder(java.util.List<georegression.struct.point.Point2D_I32> contour) Checks to see if some part of the contour touches the image border.

Methods inherited from class java.lang.Object

clone, equals, finalize, getClass, hashCode, notify, notifyAll, toString, wait, wait, wait

Field Detail

toUndistorted

protected boofcv.struct.distort.PixelTransform_F32 toUndistorted

toDistorted

protected boofcv.struct.distort.PixelTransform_F32 toDistorted

Constructor Detail

BinaryPolygonConvexDetector

public BinaryPolygonConvexDetector(int[] polygonSides,
                                   SplitMergeLineFitLoop contourToPolygon,
                                   PolygonEdgeScore differenceScore,
                                   RefinePolygonLineToImage<T> refineLine,
                                   RefinePolygonCornersToImage<T> refineCorner,
                                   double minContourFraction,
                                   double splitDistanceFraction,
                                   boolean outputClockwise,
                                   java.lang.Class<T> inputType)

Configures the detector.

Parameters:

polygonSides - Number of lines in the polygon

contourToPolygon - Fits a crude polygon to the shape's binary contour

differenceScore - Used to remove false positives by computing the difference along the polygon's edges. If null then this test is skipped.

refineLine - (Optional) Refines the polygon's lines. Set to null to skip step

refineCorner - (Optional) Refines the polygon's corners. Set to null to skip step

minContourFraction - Size of minimum contour as a fraction of the input image's width. Try 0.23

splitDistanceFraction - Number of pixels as a fraction of contour length to split a new line in SplitMergeLineFitLoop. Try 0.03

outputClockwise - If true then the order of the output polygons will be in clockwise order

inputType - Type of input image it's processing

Method Detail

setLensDistortion

public void setLensDistortion(int width,
                              int height,
                              boofcv.struct.distort.PixelTransform_F32 toUndistorted,
                              boofcv.struct.distort.PixelTransform_F32 toDistorted)

Specifies transforms which can be used to change coordinates from distorted to undistorted and the opposite coordinates. The undistorted image is never explicitly created.

WARNING: The undistorted image must have the same bounds as the distorted input image. This is because several of the bounds checks use the image shape. This are simplified greatly by this assumption.

Parameters:

width - Input image width. Used in sanity check only.

height - Input image height. Used in sanity check only.

toUndistorted - Transform from undistorted to distorted image.

toDistorted - Transform from distorted to undistorted image.

process

public void process(T gray,
                    boofcv.struct.image.ImageUInt8 binary)

Examines the undistorted gray scake input image for squares.

Parameters:

gray - Input image

touchesBorder

protected final boolean touchesBorder(java.util.List<georegression.struct.point.Point2D_I32> contour)

Checks to see if some part of the contour touches the image border. Most likely cropped

getLabeled

public boofcv.struct.image.ImageSInt32 getLabeled()

isOutputClockwise

public boolean isOutputClockwise()

getFound

public org.ddogleg.struct.FastQueue<georegression.struct.shapes.Polygon2D_F64> getFound()

getFoundContours

public java.util.List<boofcv.alg.filter.binary.Contour> getFoundContours()

getInputType

public java.lang.Class<T> getInputType()

setNumberOfSides

public void setNumberOfSides(int[] numberOfSides)

getNumberOfSides

public int[] getNumberOfSides()

setVerbose

public void setVerbose(boolean verbose)

isCheckEdgeBefore

public boolean isCheckEdgeBefore()

setCheckEdgeBefore

public void setCheckEdgeBefore(boolean checkEdgeBefore)

If set to true it will prune using polygons using their edge intensity before sub-pixel optimization. This should only be set to false if the initial edge is known to be off by a bit, like with a chessboard.

Parameters:

checkEdgeBefore - true for checking before and false for after.