Native Quality Measures

Overview

NFIQ2 computes native quality measures that assess specific characteristics of fingerprint images. These measures analyze various aspects of ridge structure, clarity, orientation, and contrast. The random forest model combines these measures to produce unified quality scores.

Native quality measures are the building blocks of NFIQ2’s quality assessment. Each measure captures a different dimension of fingerprint image quality.

Quality Measure Categories

NFIQ2 organizes quality measures into several categories:

Quality Measure Algorithms

High-level algorithms that compute one or more related measures:

namespace NFIQ2::Identifiers::QualityMeasureAlgorithms {
    extern const char FrequencyDomainAnalysis[];     // FDA algorithm
    extern const char MinutiaeCount[];               // Minutiae counting
    extern const char MinutiaeQuality[];             // Minutiae-based quality
    extern const char LocalClarity[];                // LCS algorithm
    extern const char Contrast[];                    // Image contrast (Mu)
    extern const char OrientationCertainty[];        // OCL algorithm
    extern const char OrientationFlow[];             // OF algorithm
    extern const char RegionOfInterestMean[];        // ROI mean
    extern const char RegionOfInterestCoherence[];   // ROI coherence
    extern const char RidgeValleyUniformity[];       // RVUP algorithm
}

Computing All Native Quality Measures

#include <nfiq2.hpp>

// Compute all native quality measures
auto nativeQualityMeasures = 
    NFIQ2::QualityMeasures::computeNativeQualityMeasures(fingerprintImage);

// Access individual measures by identifier
double fdaMean = nativeQualityMeasures["FrequencyDomainAnalysis_Mean"];
double lcsStdDev = nativeQualityMeasures["LocalClarity_StdDev"];
double muValue = nativeQualityMeasures["Contrast_ImageMean"];

// Get list of all available measure IDs
std::vector<std::string> measureIDs = 
    NFIQ2::QualityMeasures::getNativeQualityMeasureIDs();

Frequency Domain Analysis (FDA)

Purpose: Measures the frequency of the sinusoid following the ridge-valley structure.

Implementation Details

namespace NFIQ2::QualityMeasures {
    class FDA : public Algorithm {
    public:
        FDA(const NFIQ2::FingerprintImageData &fingerprintImage);
        std::string getName() const override;
        static std::vector<std::string> getNativeQualityMeasureIDs();
    };
}

Quality Measures Produced

Mean and StdDev
Histogram Bins

// Access FDA mean and standard deviation
double fdaMean = measures["FrequencyDomainAnalysis_Mean"];
double fdaStdDev = measures["FrequencyDomainAnalysis_StdDev"];

Mean: Average frequency domain quality across local regions
StdDev: Variability in frequency domain quality

// Access FDA histogram bins (10 bins total)
for (int i = 0; i < 10; i++) {
    std::string binKey = "FrequencyDomainAnalysis_Histogram_Bin" + std::to_string(i);
    double binValue = measures[binKey];
}

Histogram Boundaries (defined in FDA.h:18):

static double FDAHISTLIMITS[9] = { 
    0.268, 0.304, 0.33, 0.355, 0.38, 0.407, 0.44, 0.50, 1.0 
};

What FDA Measures

Ridge-valley alternation frequency consistency
Local region quality based on spectral analysis
Deviation from expected fingerprint periodicity

FDA operates on 32×32 pixel local regions, rotated to align ridges vertically (16×32 after rotation). This alignment improves frequency analysis accuracy.

Local Clarity Score (LCS)

Purpose: Measures the clarity and sharpness of ridge and valley structures.

Implementation

namespace NFIQ2::QualityMeasures {
    class LCS : public Algorithm {
    public:
        LCS(const NFIQ2::FingerprintImageData &fingerprintImage);
        static std::vector<std::string> getNativeQualityMeasureIDs();
    };
}

Quality Measures Produced

// Mean and standard deviation
double lcsMean = measures["LocalClarity_Mean"];
double lcsStdDev = measures["LocalClarity_StdDev"];

// Histogram bins (10 bins)
for (int i = 0; i < 10; i++) {
    std::string binKey = "LocalClarity_Histogram_Bin" + std::to_string(i);
    double binValue = measures[binKey];
}

Histogram Boundaries (from LCS.h:13):

static double LCSHISTLIMITS[9] = { 
    0.0, 0.70, 0.74, 0.77, 0.79, 0.81, 0.83, 0.85, 0.87 
};

What LCS Measures

Ridge edge sharpness and definition
Valley depth and clarity
Overall crispness of ridge-valley transitions

Contrast (Mu)

Purpose: Measures gray level distribution and image contrast.

Implementation

namespace NFIQ2::QualityMeasures {
    class Mu : public Algorithm {
    public:
        Mu(const NFIQ2::FingerprintImageData &fingerprintImage);
        double getSigma() const;  // Standard deviation of gray levels
    };
}

Quality Measures Produced

namespace NFIQ2::Identifiers::QualityMeasures::Contrast {
    extern const char ImageMean[];          // "Contrast_ImageMean" (MU)
    extern const char MeanOfBlockMeans[];   // "Contrast_MeanOfBlockMeans" (MMB)
}

// Access contrast measures
double mu = measures["Contrast_ImageMean"];          // Overall image mean
double mmb = measures["Contrast_MeanOfBlockMeans"]; // Average of block means

What Mu Measures

ImageMean (MU)

Arithmetic mean of all pixel gray levels in the image.

Values near 255: Image may be overexposed or blank
Values near 0: Image may be underexposed
Values near 127: Optimal contrast range

MeanOfBlockMeans (MMB)

Arithmetic mean of per-block means across the image.Provides insight into local contrast variations:

Similar to MU: Uniform contrast across image
Different from MU: Non-uniform contrast distribution

Orientation Certainty Level (OCL)

Purpose: Measures the strength of energy concentration along the dominant ridge flow orientation.

Implementation

namespace NFIQ2::QualityMeasures {
    class OCLHistogram : public Algorithm {
    public:
        OCLHistogram(const NFIQ2::FingerprintImageData &fingerprintImage);
        
        // Compute OCL for a single block
        static bool getOCLValueOfBlock(const cv::Mat &block, double &ocl);
    };
}

Quality Measures Produced

// Mean and standard deviation
double oclMean = measures["OrientationCertainty_Mean"];
double oclStdDev = measures["OrientationCertainty_StdDev"];

// Histogram bins (10 bins)
for (int i = 0; i < 10; i++) {
    std::string binKey = "OrientationCertainty_Histogram_Bin" + std::to_string(i);
    double binValue = measures[binKey];
}

Histogram Boundaries (from OCLHistogram.h:16):

static double OCLPHISTLIMITS[9] = { 
    0.337, 0.479, 0.579, 0.655, 0.716, 0.766, 0.81, 0.852, 0.898 
};

What OCL Measures

Ridge flow coherence: How consistently ridges flow in one direction
Structure quality: Strong OCL indicates well-defined ridge structure
Noise and artifacts: Low OCL may indicate smudging or poor capture

OCL is particularly sensitive to pressure variations during capture. Overcompression can create artificial high OCL values, while undercompression reduces OCL.

Orientation Flow (OF)

Purpose: Measures ridge flow continuity based on orientation differences in neighboring blocks.

Implementation

namespace NFIQ2::QualityMeasures {
    class OF : public Algorithm {
    public:
        OF(const NFIQ2::FingerprintImageData &fingerprintImage);
        
        // Minimum angle change for inclusion in quality measure
        static constexpr double angleMin { 4.0 };
    };
}

Quality Measures Produced

// Mean and standard deviation
double ofMean = measures["OrientationFlow_Mean"];
double ofStdDev = measures["OrientationFlow_StdDev"];

// Histogram bins (10 bins)
for (int i = 0; i < 10; i++) {
    std::string binKey = "OrientationFlow_Histogram_Bin" + std::to_string(i);
    double binValue = measures[binKey];
}

Histogram Boundaries (from OF.h:19):

static double OFHISTLIMITS[9] = { 
    1.715e-2, 3.5e-2, 5.57e-2, 8.1e-2, 1.15e-1, 
    1.718e-1, 2.569e-1, 4.758e-1, 7.48e-1 
};

What OF Measures

Flow continuity: Smooth orientation changes indicate good quality
Abrupt changes: Large orientation differences suggest noise or damage
Pattern consistency: Used to detect scars, creases, or dry fingers

Ridge-Valley Uniformity (RVUP)

Purpose: Measures the consistency of ridge and valley widths.

Implementation

namespace NFIQ2::QualityMeasures {
    class RVUPHistogram : public Algorithm {
    public:
        RVUPHistogram(const NFIQ2::FingerprintImageData &fingerprintImage);
    };
}

Quality Measures Produced

// Mean and standard deviation
double rvupMean = measures["RidgeValleyUniformity_Mean"];
double rvupStdDev = measures["RidgeValleyUniformity_StdDev"];

// Histogram bins (10 bins)
for (int i = 0; i < 10; i++) {
    std::string binKey = "RidgeValleyUniformity_Histogram_Bin" + std::to_string(i);
    double binValue = measures[binKey];
}

Histogram Boundaries (from RVUPHistogram.h:13):

static double RVUPHISTLIMITS[9] = { 
    0.5, 0.667, 0.8, 1.0, 1.25, 1.5, 2.0, 24.0, 30.0 
};

What RVUP Measures

Width consistency: Uniform ridge/valley widths indicate good quality
Pressure artifacts: Inconsistent widths suggest over/under pressure
Resolution issues: Extreme variations may indicate scaling problems

Minutiae-Based Measures

Purpose: Analyzes detected minutiae characteristics and distribution.

Quality Measures Produced

namespace NFIQ2::Identifiers::QualityMeasures::Minutiae {
    extern const char Count[];                          // Total minutiae count
    extern const char CountCOM[];                       // Count in center of mass
    extern const char PercentImageMean50[];             // % with good local contrast
    extern const char PercentOrientationCertainty80[];  // % with good local OCL
}

// Access minutiae measures
double minutiaeCount = measures["Minutiae_Count"];
double minutiaeCOM = measures["Minutiae_CountCOM"];
double percentMean = measures["Minutiae_PercentImageMean50"];
double percentOCL = measures["Minutiae_PercentOrientationCertainty80"];

Minutiae Measure Details

Count
CountCOM
PercentImageMean50
PercentOrientationCertainty80

Total number of minutiae detected by the FingerJet FX OSE feature extractor.Typical Ranges:

Good quality: 40-80 minutiae
Poor quality: < 20 minutiae
Possible artifacts: > 100 minutiae

Region of Interest (ROI) Measures

Purpose: Analyzes characteristics of the detected foreground region.

Quality Measures Produced

namespace NFIQ2::Identifiers::QualityMeasures::RegionOfInterest {
    extern const char Mean[];          // ROI mean gray level
    extern const char CoherenceSum[];  // Sum of coherence values
    extern const char CoherenceMean[]; // Average coherence in ROI
}

// Access ROI measures
double roiMean = measures["RegionOfInterest_Mean"];
double cohSum = measures["RegionOfInterest_CoherenceSum"];
double cohMean = measures["RegionOfInterest_CoherenceMean"];

What ROI Measures Track

Mean: Average gray level in foreground (should be well-centered in dynamic range)
CoherenceSum: Total orientation coherence across ROI blocks
CoherenceMean: Average coherence per block (CoherenceSum / number of ROI blocks)

Extracting Quality Measure Details

Getting Algorithm Objects

// Get algorithm objects (includes timing and detailed info)
auto algorithms = 
    NFIQ2::QualityMeasures::computeNativeQualityMeasureAlgorithms(fingerprintImage);

// Access as map for easier lookup
auto algorithmMap = 
    NFIQ2::QualityMeasures::getNativeQualityMeasureAlgorithms(algorithms);

// Get specific algorithm
auto fdaAlgorithm = algorithmMap["FrequencyDomainAnalysis"];
std::cout << "FDA computed in " << fdaAlgorithm->getSpeed() << " ms" << std::endl;

// Get measures from specific algorithm
auto fdaMeasures = fdaAlgorithm->getFeatures();

Performance Timing

// Get computation speeds for all algorithms
auto speeds = 
    NFIQ2::QualityMeasures::getNativeQualityMeasureAlgorithmSpeeds(algorithms);

for (const auto& [algorithmId, speedMs] : speeds) {
    std::cout << algorithmId << ": " << speedMs << " ms" << std::endl;
}

Local Region Processing

Most quality measures operate on local regions:

namespace NFIQ2::Sizes {
    // Standard local region size (ISO/IEC 29794-4:202X, Section 5.1.2)
    const unsigned int LocalRegionSquare { 32 };  // 32×32 pixels
    
    // After rotation to align ridges vertically
    const unsigned int VerticallyAlignedLocalRegionWidth { 32 };
    const unsigned int VerticallyAlignedLocalRegionHeight { 16 };
}

Why 32×32 pixels? Per ISO/IEC 29794-4:2024, Section 5.1.2: “The size for each local region shall be 32 × 32 pixels, which is sufficient to cover 2 clear ridges” at 500 PPI.

Best Practices

Select Relevant Measures

Not all applications need all measures. Choose based on your needs:

Fast screening: Use Mu, minutiae count, and ROI mean
Quality prediction: Use full set for unified quality scores
Forensic analysis: Include all measures plus histogram distributions

Monitor Measure Distributions

Track measure distributions over time to:

Detect sensor calibration drift
Identify environmental factors (temperature, humidity)
Optimize capture settings

Understand Measure Correlations

Some measures are correlated:

FDA and LCS often move together
OCL and OF reflect related aspects
Mu affects minutiae-based percentages

The random forest accounts for these correlations.

Validate Against Ground Truth

When tuning capture systems:

Compare measures to actual match performance
Identify which measures best predict problems in your specific environment
Adjust capture guidance based on measure feedback

Next Steps

Unified Quality Scores

Learn how measures combine into quality scores

Actionable Feedback

Use measures to provide capture guidance

Random Forest Model

Understand how the ML model weights measures

Quality Measures API

Complete API documentation

Get Started

Core Concepts

Guides

CLI Tool

Building from Source

Documentation Index

​Overview

​Quality Measure Categories

​Quality Measure Algorithms

​Computing All Native Quality Measures

​Frequency Domain Analysis (FDA)

​Implementation Details

​Quality Measures Produced

​What FDA Measures

​Local Clarity Score (LCS)

​Implementation

​Quality Measures Produced

​What LCS Measures

​Contrast (Mu)

​Implementation

​Quality Measures Produced

​What Mu Measures

​Orientation Certainty Level (OCL)

​Implementation

​Quality Measures Produced

​What OCL Measures

​Orientation Flow (OF)

​Implementation

​Quality Measures Produced

​What OF Measures

​Ridge-Valley Uniformity (RVUP)

​Implementation

​Quality Measures Produced

​What RVUP Measures

​Minutiae-Based Measures

​Quality Measures Produced

​Minutiae Measure Details

​Region of Interest (ROI) Measures

​Quality Measures Produced

​What ROI Measures Track

​Extracting Quality Measure Details

​Getting Algorithm Objects

​Performance Timing

​Local Region Processing

​Best Practices

​Next Steps

Unified Quality Scores

Actionable Feedback

Random Forest Model

Quality Measures API

Build docs developers (and LLMs) love

Overview

Quality Measure Categories

Quality Measure Algorithms

Computing All Native Quality Measures

Frequency Domain Analysis (FDA)

Implementation Details

Quality Measures Produced

What FDA Measures

Local Clarity Score (LCS)

Implementation

Quality Measures Produced

What LCS Measures

Contrast (Mu)

Implementation

Quality Measures Produced

What Mu Measures

Orientation Certainty Level (OCL)

Implementation

Quality Measures Produced

What OCL Measures

Orientation Flow (OF)

Implementation

Quality Measures Produced

What OF Measures

Ridge-Valley Uniformity (RVUP)

Implementation

Quality Measures Produced

What RVUP Measures

Minutiae-Based Measures

Quality Measures Produced

Minutiae Measure Details

Region of Interest (ROI) Measures

Quality Measures Produced

What ROI Measures Track

Extracting Quality Measure Details

Getting Algorithm Objects

Performance Timing

Local Region Processing

Best Practices

Next Steps