boofcv.alg.segmentation.ms

Class MergeSmallRegions<T extends boofcv.struct.image.ImageBase>

java.lang.Object

boofcv.alg.segmentation.ms.RegionMergeTree

boofcv.alg.segmentation.ms.MergeSmallRegions<T>

public class MergeSmallRegions<T extends boofcv.struct.image.ImageBase>
extends RegionMergeTree

Finds regions which are too small and merges them with a neighbor that is the most similar to it and connected. The process is repeated until there are no more regions below the size threshold. How similar two neighbors are is determined using each region's average color. Connectivity is determined using a 4-connect rule.

Nested Class Summary

Nested Classes

Modifier and Type	Class and Description
static class	MergeSmallRegions.Node Node in a graph.

Field Summary

Fields

Modifier and Type	Field and Description
protected ComputeRegionMeanColor<T>	computeColor
protected georegression.struct.point.Point2D_I32[]	connect
protected int	minimumSize
protected org.ddogleg.struct.FastQueue<MergeSmallRegions.Node>	pruneGraph
protected org.ddogleg.struct.GrowQueue_B	segmentPruneFlag
protected org.ddogleg.struct.GrowQueue_I32	segmentToPruneID

Fields inherited from class boofcv.alg.segmentation.ms.RegionMergeTree

mergeList, rootID, tmpMemberCount

Constructor Summary

Constructors

Constructor and Description
MergeSmallRegions(int minimumSize, boofcv.struct.ConnectRule rule, ComputeRegionMeanColor<T> computeColor) Constructor

Method Summary

Modifier and Type	Method and Description
protected void	adjacentBorder(boofcv.struct.image.GrayS32 pixelToRegion)
protected void	adjacentInner4(boofcv.struct.image.GrayS32 pixelToRegion)
protected void	adjacentInner8(boofcv.struct.image.GrayS32 pixelToRegion)
protected void	findAdjacentRegions(boofcv.struct.image.GrayS32 pixelToRegion) Go through each pixel in the image and examine its neighbors according to a 4-connect rule.
void	process(T image, boofcv.struct.image.GrayS32 pixelToRegion, org.ddogleg.struct.GrowQueue_I32 regionMemberCount, org.ddogleg.struct.FastQueue<float[]> regionColor) Merges together smaller regions.
protected void	selectMerge(int pruneId, org.ddogleg.struct.FastQueue<float[]> regionColor) Examine edges for the specified node and select node which it is the best match for it to merge with
void	setMinimumSize(int minimumSize)
protected boolean	setupPruneList(org.ddogleg.struct.GrowQueue_I32 regionMemberCount) Identifies which regions are to be pruned based on their member counts.

Methods inherited from class boofcv.alg.segmentation.ms.RegionMergeTree

flowIntoRootNode, initializeMerge, markMerge, performMerge, setToRootNodeNewID

Methods inherited from class java.lang.Object

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

Field Detail

minimumSize

protected int minimumSize

computeColor

protected ComputeRegionMeanColor<T extends boofcv.struct.image.ImageBase> computeColor

segmentPruneFlag

protected org.ddogleg.struct.GrowQueue_B segmentPruneFlag

segmentToPruneID

protected org.ddogleg.struct.GrowQueue_I32 segmentToPruneID

pruneGraph

protected org.ddogleg.struct.FastQueue<MergeSmallRegions.Node> pruneGraph

connect

protected georegression.struct.point.Point2D_I32[] connect

Constructor Detail

MergeSmallRegions

public MergeSmallRegions(int minimumSize,
                         boofcv.struct.ConnectRule rule,
                         ComputeRegionMeanColor<T> computeColor)

Constructor

Parameters:

minimumSize - Minimum number of pixels a region must have for it to not be pruned.

computeColor - Computes the color of each region

Method Detail

setMinimumSize

public void setMinimumSize(int minimumSize)

process

public void process(T image,
                    boofcv.struct.image.GrayS32 pixelToRegion,
                    org.ddogleg.struct.GrowQueue_I32 regionMemberCount,
                    org.ddogleg.struct.FastQueue<float[]> regionColor)

Merges together smaller regions. Segmented image, region member count, and region color are all updated.

Parameters:

image - Input image. Used to compute color of each region

pixelToRegion - (input/output) Segmented image with the ID of each region. Modified.

regionMemberCount - (input/output) Number of members in each region Modified.

regionColor - (Output) Storage for colors of each region. Will contains the color of each region on output.

setupPruneList

protected boolean setupPruneList(org.ddogleg.struct.GrowQueue_I32 regionMemberCount)

Identifies which regions are to be pruned based on their member counts. Then sets up data structures for graph and converting segment ID to prune ID.

Returns:

true If elements need to be pruned and false if not.

findAdjacentRegions

protected void findAdjacentRegions(boofcv.struct.image.GrayS32 pixelToRegion)

Go through each pixel in the image and examine its neighbors according to a 4-connect rule. If one of the pixels is in a region that is to be pruned mark them as neighbors. The image is traversed such that the number of comparisons is minimized.

adjacentInner4

protected void adjacentInner4(boofcv.struct.image.GrayS32 pixelToRegion)

adjacentInner8

protected void adjacentInner8(boofcv.struct.image.GrayS32 pixelToRegion)

adjacentBorder

protected void adjacentBorder(boofcv.struct.image.GrayS32 pixelToRegion)

selectMerge

protected void selectMerge(int pruneId,
                           org.ddogleg.struct.FastQueue<float[]> regionColor)

Examine edges for the specified node and select node which it is the best match for it to merge with

Parameters:

pruneId - The prune Id of the segment which is to be merged into another segment

regionColor - List of region colors