SegmentationMetrics

Computes classical quality metrics for assessing the quality of a segmentation relatively to a ground truth.

Access to parameter description

This algorithm allows the comparison of a segmented image with a ground truth in order to assess the accuracy of a segmentation process; for instance, a classification by machine learning.

It outputs several statistical indicators, which can be computed independently for each label of the input images, or globally to provide a unique set of scores:

The number of positive results in the input image that are defined as such in the ground truth image (TP).
The number of positive results in the input image that are not defined as such in the ground truth image (FP).
The number of negative results in the input image that are defined as such in the ground truth image (TN).
The number of negative results in the input image that are not defined as such in the ground truth image (FN).
The sensitivity, which measures the proportion of actual positives that are correctly identified as such. This metric is also called recall or true positive rate. $$Sensitivity = Recall = TPR = \frac{TP}{TP+FN}$$
The specificity, which measures the proportion of actual negatives that are correctly identified as such (true negative rate). $$ Specificity = TNR = \frac{TN}{TN+FP}$$
The precision, which measures the fraction of relevant results (true positives) among positive results (true and false positives). $$ Precision = \frac{TP}{TP+FP}$$
The accuracy, which measures the percentage of correct classifications among all results. $$ Accuracy = \frac{TP+TN}{TP+FN+TN+FP} $$
The Dice coefficient, which is a statistic used for comparing the similarity of two samples (Sørensen-Dice coefficient).This metric is also called $F_1$ score. $$ Dice = F_1 = \frac{2TP}{2TP+FP+FN} $$
The Jaccard index, also called mean Intersection-Over-Union metric. $$ Jaccard = IoU = \frac{TP}{TP+FP+FN} $$
The Panoptic Quality metric, which blends the Intersection-Over-Union scores of the $N$ analyzed classes. It is only available when computing the metrics per image. $$ PQ = \frac{ \sum_{l=1}^{N}{IoU(l)}}{TP+0.5FP+0.5FN} $$

See also

Function Syntax

This function returns a SegmentationMetricsOutput structure containing outputClassMeasurement and outputImageMeasurement.

// Output structure of the segmentationMetrics function.
struct SegmentationMetricsOutput
{
    /// The similarity results. The metrics are given for each label of the input image.
    SegmentationMetricsObjectMsr::Ptr outputClassMeasurement;
    /// The similarity results. The metrics are given globally for the input image.
    SegmentationMetricsImageMsr::Ptr outputImageMeasurement;
};

// Function prototype
SegmentationMetricsOutput
segmentationMetrics( std::shared_ptr< iolink::ImageView > inputObjectImage,
                     std::shared_ptr< iolink::ImageView > inputReferenceImage,
                     SegmentationMetrics::MetricScope metricScope,
                     SegmentationMetricsObjectMsr::Ptr outputClassMeasurement = nullptr,
                     SegmentationMetricsImageMsr::Ptr outputImageMeasurement = nullptr );

This function returns a tuple containing output_class_measurement and output_image_measurement.

// Function prototype.
segmentation_metrics(input_object_image: idt.ImageType,
                     input_reference_image: idt.ImageType,
                     metric_scope: SegmentationMetrics.MetricScope = SegmentationMetrics.MetricScope.BOTH,
                     output_class_measurement: Union[Any, None] = None,
                     output_image_measurement: Union[Any, None] = None) -> Tuple[SegmentationMetricsObjectMsr, SegmentationMetricsImageMsr]

This function returns a SegmentationMetricsOutput structure containing outputClassMeasurement and outputImageMeasurement.

/// Output structure of the SegmentationMetrics function.
public struct SegmentationMetricsOutput
{
    /// 
    /// The similarity results. The metrics are given for each label of the input image.
    /// 
    public SegmentationMetricsObjectMsr outputClassMeasurement;
    /// The similarity results. The metrics are given globally for the input image.
    public SegmentationMetricsImageMsr outputImageMeasurement;
};

// Function prototype.
public static SegmentationMetricsOutput
SegmentationMetrics( IOLink.ImageView inputObjectImage,
                     IOLink.ImageView inputReferenceImage,
                     SegmentationMetrics.MetricScope metricScope = ImageDev.SegmentationMetrics.MetricScope.BOTH,
                     SegmentationMetricsObjectMsr outputClassMeasurement = null,
                     SegmentationMetricsImageMsr outputImageMeasurement = null );

Class Syntax

// Command constructor.
SegmentationMetrics();

/// Gets the inputObjectImage parameter.
/// The input image. It can be a binary or a label image.
std::shared_ptr< iolink::ImageView > inputObjectImage() const;
/// Sets the inputObjectImage parameter.
/// The input image. It can be a binary or a label image.
void setInputObjectImage( std::shared_ptr< iolink::ImageView > inputObjectImage );

/// Gets the inputReferenceImage parameter.
/// The ground truth image. It can be a binary or a label image. It must have same shape and interpretation as the first input image.
std::shared_ptr< iolink::ImageView > inputReferenceImage() const;
/// Sets the inputReferenceImage parameter.
/// The ground truth image. It can be a binary or a label image. It must have same shape and interpretation as the first input image.
void setInputReferenceImage( std::shared_ptr< iolink::ImageView > inputReferenceImage );

/// Gets the metricScope parameter.
/// The type of output measurement to compute.
SegmentationMetrics::MetricScope metricScope() const;
/// Sets the metricScope parameter.
/// The type of output measurement to compute.
void setMetricScope( const SegmentationMetrics::MetricScope& metricScope );

/// Gets the outputClassMeasurement parameter.
/// The similarity results. The metrics are given for each label of the input image.
SegmentationMetricsObjectMsr::Ptr outputClassMeasurement() const;

/// Gets the outputImageMeasurement parameter.
/// The similarity results. The metrics are given globally for the input image.
SegmentationMetricsImageMsr::Ptr outputImageMeasurement() const;


// Method to launch the command.
void execute();

# Property of the inputObjectImage parameter.
SegmentationMetrics.input_object_image
# Property of the inputReferenceImage parameter.
SegmentationMetrics.input_reference_image
# Property of the metricScope parameter.
SegmentationMetrics.metric_scope
# Property of the outputClassMeasurement parameter.
SegmentationMetrics.output_class_measurement
# Property of the outputImageMeasurement parameter.
SegmentationMetrics.output_image_measurement

// Method to launch the command.
execute()

// Command constructor.
SegmentationMetrics()

// Property of the inputObjectImage parameter.
SegmentationMetrics.inputObjectImage
// Property of the inputReferenceImage parameter.
SegmentationMetrics.inputReferenceImage
// Property of the metricScope parameter.
SegmentationMetrics.metricScope
// Property of the outputClassMeasurement parameter.
SegmentationMetrics.outputClassMeasurement
// Property of the outputImageMeasurement parameter.
SegmentationMetrics.outputImageMeasurement

// Method to launch the command.
Execute()

Parameters

Parameter Name

Description

Type

Supported Values

Default Value

inputObjectImage

The input image. It can be a binary or a label image.

Image

Binary or Label

nullptr

inputReferenceImage

The ground truth image. It can be a binary or a label image. It must have same shape and interpretation as the first input image.

Image

Binary, Label, Grayscale or Multispectral

nullptr

metricScope

The type of output measurement to compute.

PER_CLASS	The segmentation metrics are only computed for each label of the input images. Only the object measurement output is filled.
PER_IMAGE	The segmentation metrics are computed globally on the input images. Only the image measurement output is filled.
BOTH	The segmentation metrics are computed globally and for each label of the input images. Both output measurement types are filled.

Enumeration

BOTH

outputClassMeasurement

The similarity results. The metrics are given for each label of the input image.

SegmentationMetricsObjectMsr

nullptr

outputImageMeasurement

The similarity results. The metrics are given globally for the input image.

SegmentationMetricsImageMsr

nullptr

Parameter Name

Description

Type

Supported Values

Default Value

input_object_image

The input image. It can be a binary or a label image.

image

Binary or Label

None

input_reference_image

The ground truth image. It can be a binary or a label image. It must have same shape and interpretation as the first input image.

image

Binary, Label, Grayscale or Multispectral

None

metric_scope

The type of output measurement to compute.

PER_CLASS	The segmentation metrics are only computed for each label of the input images. Only the object measurement output is filled.
PER_IMAGE	The segmentation metrics are computed globally on the input images. Only the image measurement output is filled.
BOTH	The segmentation metrics are computed globally and for each label of the input images. Both output measurement types are filled.

enumeration

BOTH

output_class_measurement

The similarity results. The metrics are given for each label of the input image.

SegmentationMetricsObjectMsr

None

output_image_measurement

The similarity results. The metrics are given globally for the input image.

SegmentationMetricsImageMsr

None

Parameter Name

Description

Type

Supported Values

Default Value

inputObjectImage

The input image. It can be a binary or a label image.

Image

Binary or Label

null

inputReferenceImage

The ground truth image. It can be a binary or a label image. It must have same shape and interpretation as the first input image.

Image

Binary, Label, Grayscale or Multispectral

null

metricScope

The type of output measurement to compute.

PER_CLASS	The segmentation metrics are only computed for each label of the input images. Only the object measurement output is filled.
PER_IMAGE	The segmentation metrics are computed globally on the input images. Only the image measurement output is filled.
BOTH	The segmentation metrics are computed globally and for each label of the input images. Both output measurement types are filled.

Enumeration

BOTH

outputClassMeasurement

The similarity results. The metrics are given for each label of the input image.

SegmentationMetricsObjectMsr

null

outputImageMeasurement

The similarity results. The metrics are given globally for the input image.

SegmentationMetricsImageMsr

null

Object Examples

auto polystyrene_sep_label = readVipImage( std::string( IMAGEDEVDATA_IMAGES_FOLDER ) + "polystyrene_sep_label.vip" );

SegmentationMetrics segmentationMetricsAlgo;
segmentationMetricsAlgo.setInputObjectImage( polystyrene_sep_label );
segmentationMetricsAlgo.setInputReferenceImage( polystyrene_sep_label );
segmentationMetricsAlgo.setMetricScope( SegmentationMetrics::MetricScope::BOTH );
segmentationMetricsAlgo.execute();

std::cout << "accuracy: " << segmentationMetricsAlgo.outputClassMeasurement()->accuracy( 0 , 0 ) ;
std::cout << "accuracy: " << segmentationMetricsAlgo.outputImageMeasurement()->accuracy( 0 ) ;

polystyrene_sep_label = imagedev.read_vip_image(imagedev_data.get_image_path("polystyrene_sep_label.vip"))

segmentation_metrics_algo = imagedev.SegmentationMetrics()
segmentation_metrics_algo.input_object_image = polystyrene_sep_label
segmentation_metrics_algo.input_reference_image = polystyrene_sep_label
segmentation_metrics_algo.metric_scope = imagedev.SegmentationMetrics.BOTH
segmentation_metrics_algo.execute()

print("accuracy: ", str(segmentation_metrics_algo.output_class_measurement.accuracy(0, 0)))
print("accuracy: ", str(segmentation_metrics_algo.output_image_measurement.accuracy(0)))

ImageView polystyrene_sep_label = Data.ReadVipImage( @"Data/images/polystyrene_sep_label.vip" );

SegmentationMetrics segmentationMetricsAlgo = new SegmentationMetrics
{
    inputObjectImage = polystyrene_sep_label,
    inputReferenceImage = polystyrene_sep_label,
    metricScope = SegmentationMetrics.MetricScope.BOTH
};
segmentationMetricsAlgo.Execute();

Console.WriteLine( "accuracy: " + segmentationMetricsAlgo.outputClassMeasurement.accuracy( 0 , 0 ) );
Console.WriteLine( "accuracy: " + segmentationMetricsAlgo.outputImageMeasurement.accuracy( 0 ) );

Function Examples

auto polystyrene_sep_label = readVipImage( std::string( IMAGEDEVDATA_IMAGES_FOLDER ) + "polystyrene_sep_label.vip" );

auto result = segmentationMetrics( polystyrene_sep_label, polystyrene_sep_label, SegmentationMetrics::MetricScope::BOTH );

std::cout << "accuracy: " << result.outputClassMeasurement->accuracy( 0 , 0 ) ;
std::cout << "accuracy: " << result.outputImageMeasurement->accuracy( 0 ) ;

polystyrene_sep_label = imagedev.read_vip_image(imagedev_data.get_image_path("polystyrene_sep_label.vip"))

result_output_class_measurement, result_output_image_measurement = imagedev.segmentation_metrics(polystyrene_sep_label, polystyrene_sep_label, imagedev.SegmentationMetrics.BOTH)

print("accuracy: ", str(result_output_class_measurement.accuracy(0, 0)))
print("accuracy: ", str(result_output_image_measurement.accuracy(0)))

ImageView polystyrene_sep_label = Data.ReadVipImage( @"Data/images/polystyrene_sep_label.vip" );

Processing.SegmentationMetricsOutput result = Processing.SegmentationMetrics( polystyrene_sep_label, polystyrene_sep_label, SegmentationMetrics.MetricScope.BOTH );

Console.WriteLine(  "accuracy: " + result.outputClassMeasurement.accuracy( 0 , 0 )  );
Console.WriteLine(  "accuracy: " + result.outputImageMeasurement.accuracy( 0 )  );

SegmentationMetrics

- Function Syntax

+ Class Syntax

Parameters

Object Examples

Function Examples

Function Syntax

Class Syntax