Hierarchical Feature Selection for Efficient Image Segmentation#

The opencv hfs module contains an efficient algorithm to segment an image. This module is implemented based on the paper Hierarchical Feature Selection for Efficient Image Segmentation, ECCV 2016. The original project was developed by Yun Liu(yun-liu/hfs).

Introduction to Hierarchical Feature Selection#

This algorithm is executed in 3 stages:

In the first stage, the algorithm uses SLIC (simple linear iterative clustering) algorithm to obtain the superpixel of the input image.

In the second stage, the algorithm view each superpixel as a node in the graph. It will calculate a feature vector for each edge of the graph. It then calculates a weight for each edge based on the feature vector and trained SVM parameters. After obtaining weight for each edge, it will exploit EGB (Efficient Graph-based Image Segmentation) algorithm to merge some nodes in the graph thus obtaining a coarser segmentation After these operations, a post process will be executed to merge regions that are smaller then a specific number of pixels into their nearby region.

In the third stage, the algorithm exploits the similar mechanism to further merge the small regions obtained in the second stage into even coarser segmentation.

After these three stages, we can obtain the final segmentation of the image. For further details about the algorithm, please refer to the original paper: Hierarchical Feature Selection for Efficient Image Segmentation, ECCV 2016

Classes#

Name

Description

class cv::hfs::HfsSegment

Class cv::hfs::HfsSegment#

#include <opencv2/hfs.hpp>

Collaboration diagram for cv::hfs::HfsSegment:

cv::hfs::HfsSegment Node1 cv::hfs::HfsSegment   + getMinRegionSizeI() + getMinRegionSizeII() + getNumSlicIter() + getSegEgbThresholdI() + getSegEgbThresholdII() + getSlicSpixelSize() + getSpatialWeight() + performSegmentCpu() + performSegmentGpu() + setMinRegionSizeI() and 6 more... + create() Node2 cv::Algorithm   + Algorithm() + ~Algorithm() + clear() + empty() + getDefaultName() + read() + save() + write() + write() + load() + loadFromString() + read() # writeFormat() Node2->Node1

cv::hfs::HfsSegment Node1 cv::hfs::HfsSegment   + getMinRegionSizeI() + getMinRegionSizeII() + getNumSlicIter() + getSegEgbThresholdI() + getSegEgbThresholdII() + getSlicSpixelSize() + getSpatialWeight() + performSegmentCpu() + performSegmentGpu() + setMinRegionSizeI() and 6 more... + create() Node2 cv::Algorithm   + Algorithm() + ~Algorithm() + clear() + empty() + getDefaultName() + read() + save() + write() + write() + load() + loadFromString() + read() # writeFormat() Node2->Node1

Public Member Functions#

Public Member Functions inherited from cv::Algorithm

Return

Name

Description

Algorithm()

~Algorithm()

void

clear()

Clears the algorithm state.

bool

empty()

Returns true if the Algorithm is empty (e.g. in the very beginning or after unsuccessful read.

String

getDefaultName()

void

read(const FileNode & fn)

Reads algorithm parameters from a file storage.

void

save(const String & filename)

void

write(
    const Ptr< FileStorage > & fs,
    const String & name = String() )

void

write(FileStorage & fs)

Stores algorithm parameters in a file storage.

void

write(
    FileStorage & fs,
    const String & name )

Static Public Member Functions#

Static Public Member Functions inherited from cv::Algorithm

Return

Name

Description

static Ptr< _Tp >

load(
    const String & filename,
    const String & objname = String() )

Loads algorithm from the file.

static Ptr< _Tp >

loadFromString(
    const String & strModel,
    const String & objname = String() )

Loads algorithm from a String.

static Ptr< _Tp >

read(const FileNode & fn)

Reads algorithm from the file node.

Additional Inherited Members#

Protected Member Functions inherited from cv::Algorithm

Return

Name

Description

void

writeFormat(FileStorage & fs)

Member Function Documentation#

getMinRegionSizeI()#

int cv::hfs::HfsSegment::getMinRegionSizeI()

Python:

cv.hfs.HfsSegment.getMinRegionSizeI() -> retval

getMinRegionSizeII()#

int cv::hfs::HfsSegment::getMinRegionSizeII()

Python:

cv.hfs.HfsSegment.getMinRegionSizeII() -> retval

getNumSlicIter()#

int cv::hfs::HfsSegment::getNumSlicIter()

Python:

cv.hfs.HfsSegment.getNumSlicIter() -> retval

getSegEgbThresholdI()#

float cv::hfs::HfsSegment::getSegEgbThresholdI()

Python:

cv.hfs.HfsSegment.getSegEgbThresholdI() -> retval

getSegEgbThresholdII()#

float cv::hfs::HfsSegment::getSegEgbThresholdII()

Python:

cv.hfs.HfsSegment.getSegEgbThresholdII() -> retval

getSlicSpixelSize()#

int cv::hfs::HfsSegment::getSlicSpixelSize()

Python:

cv.hfs.HfsSegment.getSlicSpixelSize() -> retval

getSpatialWeight()#

float cv::hfs::HfsSegment::getSpatialWeight()

Python:

cv.hfs.HfsSegment.getSpatialWeight() -> retval

performSegmentCpu()#

Mat cv::hfs::HfsSegment::performSegmentCpu(
InputArray src,
bool ifDraw = true )

Python:

cv.hfs.HfsSegment.performSegmentCpu(src[, ifDraw]) -> retval

do segmentation with cpu This method is only implemented for reference. It is highly NOT recommanded to use it.

performSegmentGpu()#

Mat cv::hfs::HfsSegment::performSegmentGpu(
InputArray src,
bool ifDraw = true )

Python:

cv.hfs.HfsSegment.performSegmentGpu(src[, ifDraw]) -> retval

do segmentation gpu

Parameters

  • src — the input image

  • ifDraw — if draw the image in the returned Mat. if this parameter is false, then the content of the returned Mat is a matrix of index, describing the region each pixel belongs to. And it’s data type is CV_16U. If this parameter is true, then the returned Mat is a segmented picture, and color of each region is the average color of all pixels in that region. And it’s data type is the same as the input image

setMinRegionSizeI()#

void cv::hfs::HfsSegment::setMinRegionSizeI(int n)

Python:

cv.hfs.HfsSegment.setMinRegionSizeI(n)

: set and get the parameter minRegionSizeI. This parameter is used in the second stage mentioned above. After the EGB segmentation, regions that have fewer pixels then this parameter will be merged into it’s adjacent region.

setMinRegionSizeII()#

void cv::hfs::HfsSegment::setMinRegionSizeII(int n)

Python:

cv.hfs.HfsSegment.setMinRegionSizeII(n)

: set and get the parameter minRegionSizeII. This parameter is used in the third stage mentioned above. It serves the same purpose as minRegionSizeI

setNumSlicIter()#

void cv::hfs::HfsSegment::setNumSlicIter(int n)

Python:

cv.hfs.HfsSegment.setNumSlicIter(n)

: set and get the parameter numSlicIter. This parameter is used in the first stage. It describes how many iteration to perform when executing SLIC.

setSegEgbThresholdI()#

void cv::hfs::HfsSegment::setSegEgbThresholdI(float c)

Python:

cv.hfs.HfsSegment.setSegEgbThresholdI(c)

: set and get the parameter segEgbThresholdI. This parameter is used in the second stage mentioned above. It is a constant used to threshold weights of the edge when merging adjacent nodes when applying EGB algorithm. The segmentation result tends to have more regions remained if this value is large and vice versa.

setSegEgbThresholdII()#

void cv::hfs::HfsSegment::setSegEgbThresholdII(float c)

Python:

cv.hfs.HfsSegment.setSegEgbThresholdII(c)

: set and get the parameter segEgbThresholdII. This parameter is used in the third stage mentioned above. It serves the same purpose as segEgbThresholdI. The segmentation result tends to have more regions remained if this value is large and vice versa.

setSlicSpixelSize()#

void cv::hfs::HfsSegment::setSlicSpixelSize(int n)

Python:

cv.hfs.HfsSegment.setSlicSpixelSize(n)

: set and get the parameter slicSpixelSize. This parameter is used in the first stage mentioned above(the SLIC stage). It describes the size of each superpixel when initializing SLIC. Every superpixel approximately has \(slicSpixelSize \times slicSpixelSize\) pixels in the beginning.

setSpatialWeight()#

void cv::hfs::HfsSegment::setSpatialWeight(float w)

Python:

cv.hfs.HfsSegment.setSpatialWeight(w)

: set and get the parameter spatialWeight. This parameter is used in the first stage mentioned above(the SLIC stage). It describes how important is the role of position when calculating the distance between each pixel and it’s center. The exact formula to calculate the distance is \(colorDistance + spatialWeight \times spatialDistance\). The segmentation result tends to have more local consistency if this value is larger.

create()#

static Ptr< HfsSegment > cv::hfs::HfsSegment::create(
int height,
int width,
float segEgbThresholdI = 0.08f,
int minRegionSizeI = 100,
float segEgbThresholdII = 0.28f,
int minRegionSizeII = 200,
float spatialWeight = 0.6f,
int slicSpixelSize = 8,
int numSlicIter = 5 )

Python:

cv.hfs.HfsSegment.create(height, width[, segEgbThresholdI[, minRegionSizeI[, segEgbThresholdII[, minRegionSizeII[, spatialWeight[, slicSpixelSize[, numSlicIter]]]]]]]) -> retval

cv.hfs.HfsSegment_create(height, width[, segEgbThresholdI[, minRegionSizeI[, segEgbThresholdII[, minRegionSizeII[, spatialWeight[, slicSpixelSize[, numSlicIter]]]]]]]) -> retval

: create a hfs object

Parameters

  • height — the height of the input image

  • width — the width of the input image

  • segEgbThresholdI — parameter segEgbThresholdI

  • minRegionSizeI — parameter minRegionSizeI

  • segEgbThresholdII — parameter segEgbThresholdII

  • minRegionSizeII — parameter minRegionSizeII

  • spatialWeight — parameter spatialWeight

  • slicSpixelSize — parameter slicSpixelSize

  • numSlicIter — parameter numSlicIter

Source file#

The documentation for this class was generated from the following file: