Biometric Quality Assessment Tool (BQAT) - CLI

BQAT is a biometric quality assessment tool for generating and analysing biometric sample quality to international standards and supporting customized metrics. It takes as input directory of biometric images/data in standard formats (e.g. wsq,png,jpg) and output both the raw quality information as well as an analysis report.

biometix.github.io

Modalities

Fingerprint

The analysis of fingerprint engine based on NIST/NFIQ2 quality features. The quality score links image quality of optical and ink 500 PPI fingerprints to operational recognition performance.

Face

The face image assessment provides metrics includes head pose, smile detection, inter-eye-distance, closed eyes, etc.

Iris

The iris sample assessment module provides various quality attributes, features, and ISO metrics.

Speech

The speech assessment provides various quality metrics, including naturalness, coloration, noisiness, etc.

pip install bqat

Example Usage:

# Print help information
bqat --help

# Print version information
bqat --version

# Run benchmarking task
bqat --benchmark

# Run samples in `data` with fingerprint mode as default
bqat --input data

# Run samples in `data` with iris mode
bqat --input data --mode iris

# Run samples in `data` with iris mode and generate EDA report
bqat --input data --mode iris --report

# Search the file with name pattern in the input folder
bqat --input data --mode iris --filename "*FINGER*"

# Search the file with specific format in the input folder
bqat --input data --mode iris --type "jp2,pgm,bmp"

# Convert the files with specific formats before scanning
bqat --input data --mode fingerprint --convert "jp2,jpeg"

# Specify the file format to convert to
bqat --input data --mode fingerprint --target wsq

# Run samples in `data` with face mode, extension function enabled, limit to 100k scan
bqat --input data --mode face --extension --limit 100000

Optional Flags

You can append optional flags as follows:

• -M, --mode (REQUIRED) Specify assessment mode (Fingerprint, Face, Iris).
• -I, --input (REQUIRED) Specify input directory
• -O, --output (OPTIONAL) Specify output csv file or directory
• -B, --benchmark (OPTIONAL) Run system benchmarking analysis
• -L, --limit (OPTIONAL) Set a limit for number of files to scan
• -F, --filename (OPTIONAL) Specify filename pattern for searching in the folder
• -S, --search (OPTIONAL) Specify file types to search within the input folder
• -C, --convert (OPTIONAL) Specify file types to convert before processing
• -T, --target (OPTIONAL) Specify target type to convert to
• -E, --extension (OPTIONAL) Enable customized extension function
• --help Show a help message

If the output or log options are not specified then the tool will use a default value.

Input & Output

Input Format

For fingerprints the tool works with image formats WSQ and PNG. For both of these formats the image will be run directly through NFIQ2. The image formats JPG and BMP are also supported but will be converted to WSQ first before being run through NFIQ2.

NFIQ2 expects images to have a resolution of at least 500 PPI. The tool will force NFIQ2 to run on images of lower resolution but the result may be inaccurate.

Output Format

The tool will produce a csv with all the quality scores generated by the engines and some additional columns.

Fingerprint

Column Name	Description
Filename	Filename of the image, including the directory path
FingerCode	NFIQ2 Output
QualityScore	NFIQ2 Output
OptionalError	NFIQ2 Output
Quantized	NFIQ2 Output
Resampled	NFIQ2 Output
UniformImage	NFIQ2 Output
EmptyImageOrContrastTooLow	NFIQ2 Output
FingerprintImageWithMinutiae	NFIQ2 Output
SufficientFingerprintForeground	NFIQ2 Output
EdgeStd	Metric to identify malformed images
Width	Width of the image in pixels
Height	Height of the image in pixels
uuid	The unique id assigned to this image

Face

Column Name	Description
Filename	Filename of the image, including the directory path
IPD	Inter-pupillary distance
Closed eye left	Bool value
Closed eye right	Bool value
Head pose yaw	Direction and degree
Head pose pitch	Direction and degree
Head pose roll	Direction and degree
Expression smile	Bool value
Face recognition confidence level	Percentage

Iris

Column Name	Description
quality	An overall quality score that leverages several statistics together
contrast	Raw score quantifying overall image contrast
sharpness	Raw score quantifying the sharpness of the image
iris_diameter	Raw diameter of the iris measured in pixels
percent_visible_iris	Percentage of visible iris area
iris_pupil_gs	Raw measure quantifying how distinguishable the boundary is between the pupil and the iris
iris_sclera_gs	Raw measure quantifying how distinguishable the boundary is between the iris and the sclera

Report

A overview statistical report on each of the column.