trainableSegmentation.metrics

Class VariationOfInformation

java.lang.Object

trainableSegmentation.metrics.Metrics

trainableSegmentation.metrics.VariationOfInformation

public class VariationOfInformation
extends Metrics

This class implements the variation of information metric, used to compare clusters or segmentations. Two implementations are given, the classic one, and a foreground-restricted version that prunes out the background pixels of the ground truth during the metric calculation.

Field Summary

Fields inherited from class trainableSegmentation.metrics.Metrics

originalLabels, proposedLabels, verbose

Constructor Summary

Constructors

Constructor and Description
VariationOfInformation(ij.ImagePlus originalLabels, ij.ImagePlus proposedLabels) Initialize variation of information metric

Method Summary

Modifier and Type	Method and Description
double	foregroundRestrictedFscore(ij.process.ShortProcessor cluster1, ij.process.ShortProcessor cluster2) Calculate the F-score of the variation of information between two clusters using the foreground restriction, i.e.
double	foregroundRestrictedFScoreN2(ij.process.ImageProcessor label, ij.process.ImageProcessor proposal, double binaryThreshold) Calculate F-score of foreground-restricted variation of information with N^2 normalization
double	foregroundRestrictedScoreAfterThinningN2(ij.process.ImageProcessor label, ij.process.ImageProcessor proposal, double binaryThreshold) Calculate foreground-restricted information theoretic score with N^2 normalization after border thinning
InformationStatistics	foregroundRestrictedStats(ij.process.ImageProcessor gt, ij.process.ImageProcessor proposal, double binaryThreshold) Get the foreground-restricted variation of information statistics (entropy values, F-score, etc)
InformationStatistics	foregroundRestrictedStats(ij.process.ShortProcessor cluster1, ij.process.ShortProcessor cluster2) Calculate the variation of information statistics (entropy values, F-score, etc) between two clusters using the foreground restriction, i.e.
double	foregroundRestrictedVI(ij.process.ShortProcessor cluster1, ij.process.ShortProcessor cluster2) Calculate the variation of information between two clusters using the foreground restriction, i.e.
double	foregroundRestrictedVIN2(ij.process.ImageProcessor label, ij.process.ImageProcessor proposal, double binaryThreshold) Calculate foreground-restricted variation of information with N^2 normalization
double	fscore(ij.process.ShortProcessor cluster1, ij.process.ShortProcessor cluster2) Calculate the F-score of the variation of information between two clusters A and B ( harmonic mean of C(A|B) and C(B|A) )
double	fScoreN2(ij.process.ImageProcessor label, ij.process.ImageProcessor proposal, double binaryThreshold) Calculate F-score of variation of information with N^2 normalization
double	getForegroundRestrictedFscore(double th) Get F-score of the foreground-restricted variation of information for a given threshold of the proposal labels.
Callable<Double>	getforegroundRestrictedFscoreConcurrent(ij.process.ImageProcessor image1, ij.process.ImageProcessor image2, double binaryThreshold) Get F-score of the foreground-restricted variation of information in a concurrent way.
ArrayList<Double>	getForegroundRestrictedFscores(double minThreshold, double maxThreshold, double stepThreshold) Get foreground-restricted F-score of the variation of information over a set of thresholds
double[]	getForegroundRestrictedGroundTruthDisagreements(ij.process.ShortProcessor cluster1, ij.process.ShortProcessor cluster2) Get foreground-restricted disagreements between prediction and ground truth labels with N^2 normalization (mergers).
double	getForegroundRestrictedMaximalFScore(double minThreshold, double maxThreshold, double stepThreshold) Get the best F-score of the foreground-restricted variation of information over a set of thresholds
double	getForegroundRestrictedMetricValue(double binaryThreshold) Get foreground-restricted variation of information between original and proposed labels for a given threshold
double[]	getForegroundRestrictedPredictionDisagreements(ij.process.ShortProcessor cluster1, ij.process.ShortProcessor cluster2) Get foreground-restricted disagreements between ground truth and prediction labels with N^2 normalization (splits).
double	getForegroundRestrictedScoreAfterThinning(double th) Get foreground-restricted information theoretic score after border thinning for a given threshold of the proposal labels.
Callable<Double>	getforegroundRestrictedScoreAfterThinningConcurrent(ij.process.ImageProcessor image1, ij.process.ImageProcessor image2, double binaryThreshold) Get foreground-restricted information theoretic score after border thinning in a concurrent way.
ArrayList<Double>	getForegroundRestrictedScoresAfterThinning(double minThreshold, double maxThreshold, double stepThreshold) Get foreground-restricted information theoretic score after thinning over a set of thresholds
InformationStatistics[]	getForegroundRestrictedStatsPerSlice(double binaryThreshold) Get the foreground-restricted variation of information statistics (entropy values, F-score, etc) per slice and for a given threshold value.
Callable<Double>	getForegroundRestrictedVIConcurrent(ij.process.ImageProcessor image1, ij.process.ImageProcessor image2, double binaryThreshold) Calculate foreground-restricted variation of information between two images in a concurrent way (to be submitted to an Executor Service).
double	getFscore(double th) Get F-score of the variation of information for a given threshold of the proposal labels.
Callable<Double>	getFscoreConcurrent(ij.process.ImageProcessor image1, ij.process.ImageProcessor image2, double binaryThreshold) Get F-score of the variation of information in a concurrent way.
ArrayList<Double>	getFscores(double minThreshold, double maxThreshold, double stepThreshold) Get F-score of the variation of information over a set of thresholds
double	getMaximalFScore(double minThreshold, double maxThreshold, double stepThreshold) Get the best F-score of the variation of information over a set of thresholds
double	getMaximalVInfoAfterThinning(double minThreshold, double maxThreshold, double stepThreshold) Get the best V_Info after thinning over a set of thresholds.
double	getMetricValue(double binaryThreshold) Get variation of information between original and proposed labels for a given threshold
double	getMinimumForegroundRestrictedMetricValue(double minThreshold, double maxThreshold, double stepThreshold) Get the minimum foreground-restricted metric value over a set of thresholds
double	getVI(ij.process.ShortProcessor cluster1, ij.process.ShortProcessor cluster2) Calculate the variation of information between two clusters (N^2 normalization)
Callable<Double>	getVIConcurrent(ij.process.ImageProcessor image1, ij.process.ImageProcessor image2, double binaryThreshold) Calculate variation of information between two images in a concurrent way (to be submitted to an Executor Service).
double	variationOfInformationN2(ij.process.ImageProcessor label, ij.process.ImageProcessor proposal, double binaryThreshold) Calculate variation of information with N^2 normalization

Methods inherited from class trainableSegmentation.metrics.Metrics

getMinimumMetricValue, setVerboseMode

Methods inherited from class java.lang.Object

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

Constructor Detail

VariationOfInformation

public VariationOfInformation(ij.ImagePlus originalLabels,
                              ij.ImagePlus proposedLabels)

Initialize variation of information metric

Parameters:

originalLabels - original labels (single 2D image or stack)

proposedLabels - proposed new labels (single 2D image or stack of the same as as the original labels)

Method Detail

getMetricValue

public double getMetricValue(double binaryThreshold)

Get variation of information between original and proposed labels for a given threshold

Specified by:

getMetricValue in class Metrics

Parameters:

binaryThreshold - threshold value to binarize labels

Returns:

metric value

getForegroundRestrictedMetricValue

public double getForegroundRestrictedMetricValue(double binaryThreshold)

Get foreground-restricted variation of information between original and proposed labels for a given threshold

Parameters:

binaryThreshold - threshold value to binarize labels

Returns:

foreground-restricted metric value

getMinimumForegroundRestrictedMetricValue

public double getMinimumForegroundRestrictedMetricValue(double minThreshold,
                                                        double maxThreshold,
                                                        double stepThreshold)

Get the minimum foreground-restricted metric value over a set of thresholds

Parameters:

minThreshold - minimum threshold value to binarize the input images

maxThreshold - maximum threshold value to binarize the input images

stepThreshold - threshold step value to use during binarization

Returns:

minimum foreground-restricted value of the metric

getVIConcurrent

public Callable<Double> getVIConcurrent(ij.process.ImageProcessor image1,
                                        ij.process.ImageProcessor image2,
                                        double binaryThreshold)

Calculate variation of information between two images in a concurrent way (to be submitted to an Executor Service). Both images are binarized.

Parameters:

image1 - first image

image2 - second image

binaryThreshold - threshold to apply to both images

Returns:

variation of information

getForegroundRestrictedVIConcurrent

public Callable<Double> getForegroundRestrictedVIConcurrent(ij.process.ImageProcessor image1,
                                                            ij.process.ImageProcessor image2,
                                                            double binaryThreshold)

Calculate foreground-restricted variation of information between two images in a concurrent way (to be submitted to an Executor Service). Both images are binarized.

Parameters:

image1 - first image

image2 - second image

binaryThreshold - threshold to apply to both images

Returns:

foreground-restricted variation of information

variationOfInformationN2

public double variationOfInformationN2(ij.process.ImageProcessor label,
                                       ij.process.ImageProcessor proposal,
                                       double binaryThreshold)

Calculate variation of information with N^2 normalization

Parameters:

label - original labels

proposal - proposed labels (usually a probability image to be thresholded)

binaryThreshold - threshold value to binarize proposal

Returns:

variation of information between original labels and the binarized proposal

foregroundRestrictedVIN2

public double foregroundRestrictedVIN2(ij.process.ImageProcessor label,
                                       ij.process.ImageProcessor proposal,
                                       double binaryThreshold)

Calculate foreground-restricted variation of information with N^2 normalization

Parameters:

label - original labels

proposal - proposed labels (usually a probability image to be thresholded)

binaryThreshold - threshold value to binarize proposal

Returns:

foreground-restricted variation of information between original labels and the binarized proposal

getVI

public double getVI(ij.process.ShortProcessor cluster1,
                    ij.process.ShortProcessor cluster2)

Calculate the variation of information between two clusters (N^2 normalization)

Parameters:

cluster1 - labels of cluster 1 (ground truth)

cluster2 - labels of cluster 2 (proposal)

Returns:

variation of information value

foregroundRestrictedVI

public double foregroundRestrictedVI(ij.process.ShortProcessor cluster1,
                                     ij.process.ShortProcessor cluster2)

Calculate the variation of information between two clusters using the foreground restriction, i.e. pruning out the zero component in the labeling (un-assigned "out" space)

Parameters:

cluster1 - labels of cluster 1 (ground truth)

cluster2 - labels of cluster 2 (proposal)

Returns:

foreground-restricted variation of information value

getForegroundRestrictedGroundTruthDisagreements

public double[] getForegroundRestrictedGroundTruthDisagreements(ij.process.ShortProcessor cluster1,
                                                                ij.process.ShortProcessor cluster2)

Get foreground-restricted disagreements between prediction and ground truth labels with N^2 normalization (mergers).

Parameters:

cluster1 - ground truth cluster

cluster2 - proposed cluster

Returns:

normalized foreground-restricted mergers per label

getForegroundRestrictedPredictionDisagreements

public double[] getForegroundRestrictedPredictionDisagreements(ij.process.ShortProcessor cluster1,
                                                               ij.process.ShortProcessor cluster2)

Get foreground-restricted disagreements between ground truth and prediction labels with N^2 normalization (splits).

Parameters:

cluster1 - ground truth cluster

cluster2 - proposed cluster

Returns:

normalized foreground-restricted splits per label

getForegroundRestrictedMaximalFScore

public double getForegroundRestrictedMaximalFScore(double minThreshold,
                                                   double maxThreshold,
                                                   double stepThreshold)

Get the best F-score of the foreground-restricted variation of information over a set of thresholds

Parameters:

minThreshold - minimum threshold value to binarize the input images

maxThreshold - maximum threshold value to binarize the input images

stepThreshold - threshold step value to use during binarization

Returns:

maximal F-score of the foreground-restricted variation of information

getMaximalVInfoAfterThinning

public double getMaximalVInfoAfterThinning(double minThreshold,
                                           double maxThreshold,
                                           double stepThreshold)

Get the best V_Info after thinning over a set of thresholds. Note: the background pixels of the ground truth are pruned out in the calculations and the background pixels of the proposal are thinned to a 1-pixel width line using the classic watershed algorithm.

Parameters:

minThreshold - minimum threshold value to binarize the input images

maxThreshold - maximum threshold value to binarize the input images

stepThreshold - threshold step value to use during binarization

Returns:

maximal Information theoretic score after thinning

getMaximalFScore

public double getMaximalFScore(double minThreshold,
                               double maxThreshold,
                               double stepThreshold)

Get the best F-score of the variation of information over a set of thresholds

Parameters:

minThreshold - minimum threshold value to binarize the input images

maxThreshold - maximum threshold value to binarize the input images

stepThreshold - threshold step value to use during binarization

Returns:

maximal F-score of the variation of information

getForegroundRestrictedFscores

public ArrayList<Double> getForegroundRestrictedFscores(double minThreshold,
                                                        double maxThreshold,
                                                        double stepThreshold)

Get foreground-restricted F-score of the variation of information over a set of thresholds

Parameters:

minThreshold - minimum threshold value to check (included)

maxThreshold - maximum threshold value to check (included)

stepThreshold - step threshold value

Returns:

list with the F-score value of each threshold

getForegroundRestrictedScoresAfterThinning

public ArrayList<Double> getForegroundRestrictedScoresAfterThinning(double minThreshold,
                                                                    double maxThreshold,
                                                                    double stepThreshold)

Get foreground-restricted information theoretic score after thinning over a set of thresholds

Parameters:

minThreshold - minimum threshold value to check (included)

maxThreshold - maximum threshold value to check (included)

stepThreshold - step threshold value

Returns:

list with the F-score value of each threshold

getFscores

public ArrayList<Double> getFscores(double minThreshold,
                                    double maxThreshold,
                                    double stepThreshold)

Get F-score of the variation of information over a set of thresholds

Parameters:

minThreshold - minimum threshold value to check (included)

maxThreshold - maximum threshold value to check (included)

stepThreshold - step threshold value

Returns:

list with the F-score value of each threshold

getForegroundRestrictedFscore

public double getForegroundRestrictedFscore(double th)

Get F-score of the foreground-restricted variation of information for a given threshold of the proposal labels. Done in 2d, for stacks, the average F-score is calculated.

Parameters:

th - threshold value to binarize proposal

Returns:

F-score of the foreground-restricted variation of information

getForegroundRestrictedScoreAfterThinning

public double getForegroundRestrictedScoreAfterThinning(double th)

Get foreground-restricted information theoretic score after border thinning for a given threshold of the proposal labels. Done in 2d, for stacks, the per-slice average score is calculated.

Parameters:

th - threshold value to binarize proposal

Returns:

foreground-restricted information theoretic score after thinning

getForegroundRestrictedStatsPerSlice

public InformationStatistics[] getForegroundRestrictedStatsPerSlice(double binaryThreshold)

Get the foreground-restricted variation of information statistics (entropy values, F-score, etc) per slice and for a given threshold value.

Parameters:

binaryThreshold - threshold value to binarize proposal ([0 1])

Returns:

foreground-restricted information statistics per slice

foregroundRestrictedStats

public InformationStatistics foregroundRestrictedStats(ij.process.ImageProcessor gt,
                                                       ij.process.ImageProcessor proposal,
                                                       double binaryThreshold)

Get the foreground-restricted variation of information statistics (entropy values, F-score, etc)

Parameters:

gt - 2D image with the original labels

proposal - 2D image with the proposed labels

binaryThreshold - threshold value to binarize the input images

Returns:

variation of information statistics

getFscore

public double getFscore(double th)

Get F-score of the variation of information for a given threshold of the proposal labels. Done in 2d, for stacks, the average F-score is calculated.

Parameters:

th - threshold value to binarize proposal

Returns:

F-score of the variation of information

getFscoreConcurrent

public Callable<Double> getFscoreConcurrent(ij.process.ImageProcessor image1,
                                            ij.process.ImageProcessor image2,
                                            double binaryThreshold)

Get F-score of the variation of information in a concurrent way. To be submitted to an ExecutorService.

Parameters:

image1 - ground truth (usually binary labels)

image2 - proposed labels (usually a probability map to binarize)

binaryThreshold - threshold value to binarize proposal

Returns:

F-score of the variation of information (N^2 normalization)

getforegroundRestrictedFscoreConcurrent

public Callable<Double> getforegroundRestrictedFscoreConcurrent(ij.process.ImageProcessor image1,
                                                                ij.process.ImageProcessor image2,
                                                                double binaryThreshold)

Get F-score of the foreground-restricted variation of information in a concurrent way. To be submitted to an ExecutorService.

Parameters:

image1 - ground truth (usually binary labels)

image2 - proposed labels (usually a probability map to binarize)

binaryThreshold - threshold value to binarize proposal

Returns:

F-score of the variation of information (foreground-restricted and N^2 normalization)

getforegroundRestrictedScoreAfterThinningConcurrent

public Callable<Double> getforegroundRestrictedScoreAfterThinningConcurrent(ij.process.ImageProcessor image1,
                                                                            ij.process.ImageProcessor image2,
                                                                            double binaryThreshold)

Get foreground-restricted information theoretic score after border thinning in a concurrent way. To be submitted to an ExecutorService.

Parameters:

image1 - ground truth (usually binary labels)

image2 - proposed labels (usually a probability map to binarize)

binaryThreshold - threshold value to binarize proposal

Returns:

information theoretic score (foreground-restricted, proposal border thinning and N^2 normalization)

fScoreN2

public double fScoreN2(ij.process.ImageProcessor label,
                       ij.process.ImageProcessor proposal,
                       double binaryThreshold)

Calculate F-score of variation of information with N^2 normalization

Parameters:

label - original labels

proposal - proposed labels (usually a probability image to be thresholded)

binaryThreshold - threshold value to binarize proposal

Returns:

F-score of the variation of information between original labels and the binarized proposal

foregroundRestrictedFScoreN2

public double foregroundRestrictedFScoreN2(ij.process.ImageProcessor label,
                                           ij.process.ImageProcessor proposal,
                                           double binaryThreshold)

Calculate F-score of foreground-restricted variation of information with N^2 normalization

Parameters:

label - original labels

proposal - proposed labels (usually a probability image to be thresholded)

binaryThreshold - threshold value to binarize proposal

Returns:

F-score of the variation of information between original labels and the binarized proposal

foregroundRestrictedScoreAfterThinningN2

public double foregroundRestrictedScoreAfterThinningN2(ij.process.ImageProcessor label,
                                                       ij.process.ImageProcessor proposal,
                                                       double binaryThreshold)

Calculate foreground-restricted information theoretic score with N^2 normalization after border thinning

Parameters:

label - original labels

proposal - proposed labels (usually a probability image to be thresholded)

binaryThreshold - threshold value to binarize proposal

Returns:

information theoretic score between original labels and the binarized proposal

foregroundRestrictedFscore

public double foregroundRestrictedFscore(ij.process.ShortProcessor cluster1,
                                         ij.process.ShortProcessor cluster2)

Calculate the F-score of the variation of information between two clusters using the foreground restriction, i.e. pruning out the zero component in the labeling (un-assigned "out" space)

Parameters:

cluster1 - labels of cluster 1 (ground truth)

cluster2 - labels of cluster 2 (proposal)

Returns:

F-score of the variation of information

foregroundRestrictedStats

public InformationStatistics foregroundRestrictedStats(ij.process.ShortProcessor cluster1,
                                                       ij.process.ShortProcessor cluster2)

Calculate the variation of information statistics (entropy values, F-score, etc) between two clusters using the foreground restriction, i.e. pruning out the zero component in the labeling (un-assigned "out" space)

Parameters:

cluster1 - labels of cluster 1 (ground truth)

cluster2 - labels of cluster 2 (proposal)

Returns:

variation of information statistics

fscore

public double fscore(ij.process.ShortProcessor cluster1,
                     ij.process.ShortProcessor cluster2)

Calculate the F-score of the variation of information between two clusters A and B ( harmonic mean of C(A|B) and C(B|A) )

Parameters:

cluster1 - labels of cluster 1 (ground truth)

cluster2 - labels of cluster 2 (proposal)

Returns:

F-score of the variation of information