Package boofcv.alg.filter.binary

Class ThresholdBlockMinMax<T extends ImageGray<T>,I extends ImageInterleaved<I>>

java.lang.Object

boofcv.alg.filter.binary.ThresholdBlockMinMax<T,I>

All Implemented Interfaces:

ThresholdBlock.BlockProcessor<T,I>

Direct Known Subclasses:

ThresholdBlockMinMax_F32, ThresholdBlockMinMax_U8

public abstract class ThresholdBlockMinMax<T extends ImageGray<T>,I extends ImageInterleaved<I>>
extends Object
implements ThresholdBlock.BlockProcessor<T,I>

Applies a threshold to an image by computing the min and max values in a regular grid across the image. When thresholding all the pixels inside a box (grid element) the min max values is found in the surrounding 3x3 grid region. If the difference between min and max is ≤ textureThreshold then it will be marked as one, since it is considered a textureless region. Otherwise the pixel threshold is set to (min+max)/2.

This thresholding strategy is designed to quickly detect shapes with nearly uniform values both inside the image and along the image border, with locally variable intensity values. The image border is particularly problematic since there are no neighboring pixels outside the image from which to compute a local threshold. This is why if a region is considered textureless it is marked as 1.

The min-max values inside a local 3x3 grid region is used to reduce the adverse affects of using a grid. Ideally a local region around each pixel would be used, but this is expensive to compute. Since a grid is used instead of a pixel local region boundary conditions can be an issue. For example, consider a black square in the image, if the grid just happens to lie on this black square perfectly then if you look at only a single grid element it will be considered textureless and the edge lost. This problem isn't an issue if you consder a local 3x3 region of blocks.

The size each block in the grid in pixels is adjusted depending on image size. This is done to minimize "squares" in the upper image boundaries from having many more pixels than other blocks.

The block based approach used here was inspired by a high level description found in AprilTags.

Field Summary

Fields

Modifier and Type	Field	Description
protected int	blockHeight
protected int	blockWidth
protected boolean	down
protected double	minimumSpread
protected boolean	thresholdFromLocalBlocks

Constructor Summary

Constructors

Modifier	Constructor	Description
protected	ThresholdBlockMinMax(double minimumSpread, boolean down)	Configures the detector

Method Summary

Modifier and Type	Method	Description
void	init(int blockWidth, int blockHeight, boolean thresholdFromLocalBlocks)

Methods inherited from class java.lang.Object

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

Methods inherited from interface boofcv.alg.filter.binary.ThresholdBlock.BlockProcessor

computeBlockStatistics, copy, createStats, thresholdBlock

Field Details

blockWidth

protected int blockWidth

blockHeight

protected int blockHeight

thresholdFromLocalBlocks

protected boolean thresholdFromLocalBlocks

minimumSpread

protected double minimumSpread

down

protected boolean down

Constructor Details

ThresholdBlockMinMax

protected ThresholdBlockMinMax(double minimumSpread,
 boolean down)

Configures the detector

Parameters:

minimumSpread - If the difference between min max is less than or equal to this value then it is considered textureless. Set to ≤ -1 to disable.

Method Details

init

public void init(int blockWidth,
 int blockHeight,
 boolean thresholdFromLocalBlocks)

Specified by:

init in interface ThresholdBlock.BlockProcessor<T extends ImageGray<T>,I extends ImageInterleaved<I>>