Feature Extraction

Feature extraction is the process of extracting unique features from a fingerprint image. It outputs a finger template to be used for one-to-one comparison or one-to-many search.

This component includes several complementary features:

  • Minutia detection – This is the core algorithm for fingerprint feature extraction. Minutiae are the most interoperable features and their encoding is precisely defined in ANSI and ISO standards.

  • Minutia encoding – Extracts a local signature for each minutia to improve overall matching performance.

  • Finger encoding – Extracts a signature of the whole fingerprint to improve the recognition performance. To be used in addition to the minutia detection.

Hint

For finger encoding, ensure that the fingerprint image is large enough (above 200x200px at 500dpi) and well captured (correct pose).

Important

Finger encoding requires the finger aligner model to be specified.

Minutiae

Minutiae are points located at the places in the fingerprint image where friction ridges end or split into two ridges. Describing a fingerprint in terms of the location and direction of these ridge endings and bifurcations provides sufficient information to reliably determine whether two fingerprint records are from the same finger. A typical good-quality fingerprint image contains about 20-70 minutiae points; the actual number depends on the size of the sensor surface and how the user places his or her finger on the sensor. Each minutia has a “type” associated with it. There are two major types of minutiae:

  • Ridge endings

  • Ridge bifurcations

../_images/minutiae.jpg

Minutia placement

To ensure the best interoperability with other systems, the minutia placement is as follow:

  • A minutia on a ridge ending is defined as the valley skeleton bifurcation point.

  • A minutia on a ridge bifurcation is defined as the ridge skeleton bifurcation point.

../_images/MinutiaPlacementOnARidgeEnding.jpg ../_images/MinutiaPlacementOnARidgeBifurcation.jpg

Minutia direction

The minutia angle is measured increasing counter-clockwise starting from the horizontal axis to the right. The angle granularity is 1.40625 (360/256) degrees.

Proprietary information

Minutia embedding

A minutia embedding is a local signature representing the neighboorhood of a minutia. It is generated using a deep learning model (see AI Models for details). It is complementary to the positions and angles of the minutiae described above.

Those signatures are stored in the proprietary section of a finger template at a cost of 16 bytes per minutia.

Finger embedding

A finger embedding is a global signature of a fingerprint. It is generated using a deep learning model (see AI Models for details). This signature contains texture details of the whole fingerprint which is complementary with both the minutiae positions, angles and textures described above.

This signature is stored in the proprietary section of a finger template at a cost of 128 bytes.

Finger Template

A finger template is a representation of a fingerprint using the fundamental notion of minutiae and optionally proprietary information. It is the result of the feature extraction process applied to a finger image.

A FingerTemplate object contains the following elements:

  • Minutia data

  • Proprietary fingerprint information such as minutia embeddings and/or finger embedding

  • Image information

    • Height and width

    • Horizontal and vertical resolution

    • Impression type

    • Position of the finger

    • Image quality

  • Capture device information

    • Certification

    • Technology

    • Type identifier (vendor specific)

    • Vendor identifier (registered by IBIA)

It can be stored in a supported standard format for further usage. See Supported Formats for details.

Note

Several templates from one or more fingers can be grouped into a single FingerTemplateRecord Class.

Template quality

id3 Finger SDK computes a quality for each created template. This value between 0 and 100 aims at predicting the ability of a template to perform well at matching. A very high template quality means that error rates during verification will be as low as possible. On the contrary, a low template quality will induce errors during verification and more precisely false rejections: a genuine person will possibly be rejected by the matching algorithm because of the poor quality of the presented template. This is why it is important to check template quality in an enrollment process.

Example

The example below demonstrates how to generate a FingerTemplate from a FingerImage:

# initialize FingerExtractor
finger_extractor = FingerExtractor(
    finger_aligner_model=FingerModel.FINGER_ALIGNER_1A,
    finger_encoder_model=FingerModel.FINGER_ENCODER_1A,
    thread_count=4
)

# load an image from a file.
image = FingerImage.from_file("data/finger1_1.bmp", PixelFormat.GRAYSCALE_8_BITS)
image.set_resolution(500)

# extract the template from the fingerprint image.
tic = time.perf_counter()
finger_template = finger_extractor.create_template(image)
toc = time.perf_counter()
print(f"Template extracted in {1000*(toc-tic):0.1f} ms")
print(f"Template quality: {finger_template.quality}")

# export the finger template to a buffer in ISO/IEC 19794-2:2011 format.
template_data = finger_template.to_buffer(FingerTemplateFormat.ISO19794_2_2011)
print(" ".join("{:02x}".format(x) for x in template_data))

# dispose all resources
del finger_extractor

See also