Server SDK for fingerprint recognition

Neurodactyl SDK helps system integrators and software providers to implement biometric fingerprint recognition - verification or identification - into their products, systems and services. The SDK is based on deep learning technologies and matches any types of fingerprints between each other: flats, rolls, latents and photo images, creating fixed-size and compatible template for any type of fingerprints. Recognition algorithm of Neurodactyl server SDK has archived world's top tier accuracy in NIST PFT III benchmarks.
  • 1
    Fingerprints detection
    The SDK detects fingerprints on photo images and scans (250 dpi and higher). The detector returns 2 landmarks and bounding boxes for each detected fingerprint. Number of fingerprints/fingers on an image is not limited. Left/right hand detection based on fingerprint analysis (optional).
  • 2
    Biometric template extraction
    The SDK converts an image into compact descriptor, describing unique features of a fingerprint. All templates extracted from photo images and scans are compatible and have standard size - 512 bytes.
  • 3
    The SDK compares templates against each other in different modes: 1:1, 1:N, M:N (batch mode) and returns similarity score (native values equal to -logFAR and %).
  • 4
    Images decompression
    The SDK has decompression feature including WSQ format.
  • 5
All processing can be performed on CPU or GPU. In terms of cost-efficiency GPU processing outperforms CPU. Neurodactyl SDK automatically utilizes all available resources of CPU or GPU without running parallel threads. You can use batch mode on GPU for detection and extraction, and batch on CPU - for matching. Batch mode accelerates processing and provides better throughput. Consult our specialists to choose optimum batches for your hardware.
  • 1
    Biometric template
    Size of biometric template is constant: 512 bytes in memory (serialized template is 546 bytes). All enrolled templates are stored in RAM, so required size of RAM depends on the size of your enrollment database.
  • 2
    Detection speed
    Detection can be performed on CPU or GPU. Average detection speed: 10-20 ms per 1 image on GPU and 70-100 ms - on CPU. Average detection speed in batch on GPU: 2-4 ms per 1 image.
  • 3
    Template extraction speed
    Template extraction can be performed on CPU or GPU. Average extraction speed: 50 ms per 1 image on GPU and 250-500 ms - on CPU. Average extraction speed in batch on GPU: 5-10 ms per 1 image.
  • 4
    Matching speed
    100 million matching operation per 1 s on one modern CPU. In batch mode: up to 1,5 billion matches per 1 s. Benchmarks for CPU models:
    • Intel Xeon Gold 6256 - 538 ms for 100 million matches; 1,7 s for 3,2 billion matches in batch.
    • Intel Core i9 10900X - 925 ms for 100 million matches; 3 s for 3 billion matches in batch.
At least 6 memory channels, avx512 instructions
Measured on GPUs: GTX 1080Ti, RTX 2080Ti, RTX 3090, image resolution=800x800 pixels, 500 dpi
Numbers for batch mode are valid for M:N matching and not equal to 1:N matching made M times. Indicated numbers are calculated for 32:100M matches (batch=32).
Measured on GPUs: GTX 1080Ti, RTX 2080Ti, RTX 3090, batch size=32 images, image resolution=800x800 pixels, 500 dpi
Measured on GPUs: GTX 1080Ti, RTX 2080Ti, RTX 3090. Average number for mentioned GPU models.
The range is indicated for following CPU models: Intel Core i9-10900X, AMD Ryzen Threadripper 2920X, Intel Core i7-8700K, Intel Core i9-11900K
The range is indicated for following GPU models: GTX 1080Ti, RTX 2080Ti, RTX 3090. Batch size=32 images.
Interfaces: C++, C# with code examples. Neurodactyl REST API platform is available for Neurodactyl SDK.

Platforms: Windows 7 or later (amd64), Linux (amd64)
Minimum HW requirements: CPU Intel or AMD with AVX2 instructions, 8 GB RAM, 4 GB free space on a drive.

HW requirements for a particular use case should be calculated for a project and depend on: size of enrollment database, number of incoming images per 1 s, types and resolutions of incoming images.

Supported GPU: NVidia GPUs starting from Pascal architecture or later, at least 6 GB RAM.

Image requirements:
  • For scans - 250 dpi and higher, rolls, flats and latents are supported.
  • For photo images - quality of images (resolution and sharpness) must provide visible fingerprints patterns. Size of a phalanx must be at least 200 pixels. We recommend to use Neurodactyl Mobile capture SDK for image acquisition with mobile phones.
Supported image formats: png, jpeg, bmp, wsq and others.

Nuerodactyl SDK is licensed per:
  • Number of enrolled images (size of N in the matching gallery)
  • Number of GPUs
  • Number of servers/machines

Price calculation can be done based on following project information:
  • Size of the enrollment database
  • Workload: number of incoming images for detection and template extraction per 1 s
  • Types of incoming images: photo/scans, resolution
  • CPU or GPU is used for processing
If you want to know more about licensing and pricing, please contact us
SDK evaluation request
If you want to get evaluation license, please fill the form below or send us an email to: