Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Effectiveness of learning-based image codecs on fingerprint storage

Daniele Mari, Saverio Cavasin, Simone Milani, Mauro Conti

TL;DR

This work investigates learning-based image codecs for biometric fingerprint storage, evaluating pretrained and fingerprint-adapted codecs (Finger-MSH) against JPEG2000 on the CASIA-FingerprintV5 dataset. It leverages a mean-and-scale hyperprior framework and Finger-MSH adaptations to quantify rate-distortion performance and minutiae preservation, demonstrating significant BD-rate savings and superior locus/quality metrics at comparable bitrates. The results indicate strong potential for learned codecs in biometric data compression, with pretrained models offering robust performance even when trained on natural images, and highlight opportunities to further optimize minutiae preservation through task-specific losses and broader biometric testing. Overall, the findings support adopting JPEG-AI-style learned codecs for fingerprint and related biometric storage, with practical implications for storage efficiency and automatic identification reliability.

Abstract

The success of learning-based coding techniques and the development of learning-based image coding standards, such as JPEG-AI, point towards the adoption of such solutions in different fields, including the storage of biometric data, like fingerprints. However, the peculiar nature of learning-based compression artifacts poses several issues concerning their impact and effectiveness on extracting biometric features and landmarks, e.g., minutiae. This problem is utterly stressed by the fact that most models are trained on natural color images, whose characteristics are very different from usual biometric images, e.g, fingerprint or iris pictures. As a matter of fact, these issues are deemed to be accurately questioned and investigated, being such analysis still largely unexplored. This study represents the first investigation about the adaptability of learning-based image codecs in the storage of fingerprint images by measuring its impact on the extraction and characterization of minutiae. Experimental results show that at a fixed rate point, learned solutions considerably outperform previous fingerprint coding standards, like JPEG2000, both in terms of distortion and minutiae preservation. Indeed, experimental results prove that the peculiarities of learned compression artifacts do not prevent automatic fingerprint identification (since minutiae types and locations are not significantly altered), nor do compromise image quality for human visual inspection (as they gain in terms of BD rate and PSNR of 47.8% and +3.97dB respectively).

Effectiveness of learning-based image codecs on fingerprint storage

TL;DR

This work investigates learning-based image codecs for biometric fingerprint storage, evaluating pretrained and fingerprint-adapted codecs (Finger-MSH) against JPEG2000 on the CASIA-FingerprintV5 dataset. It leverages a mean-and-scale hyperprior framework and Finger-MSH adaptations to quantify rate-distortion performance and minutiae preservation, demonstrating significant BD-rate savings and superior locus/quality metrics at comparable bitrates. The results indicate strong potential for learned codecs in biometric data compression, with pretrained models offering robust performance even when trained on natural images, and highlight opportunities to further optimize minutiae preservation through task-specific losses and broader biometric testing. Overall, the findings support adopting JPEG-AI-style learned codecs for fingerprint and related biometric storage, with practical implications for storage efficiency and automatic identification reliability.

Abstract

The success of learning-based coding techniques and the development of learning-based image coding standards, such as JPEG-AI, point towards the adoption of such solutions in different fields, including the storage of biometric data, like fingerprints. However, the peculiar nature of learning-based compression artifacts poses several issues concerning their impact and effectiveness on extracting biometric features and landmarks, e.g., minutiae. This problem is utterly stressed by the fact that most models are trained on natural color images, whose characteristics are very different from usual biometric images, e.g, fingerprint or iris pictures. As a matter of fact, these issues are deemed to be accurately questioned and investigated, being such analysis still largely unexplored. This study represents the first investigation about the adaptability of learning-based image codecs in the storage of fingerprint images by measuring its impact on the extraction and characterization of minutiae. Experimental results show that at a fixed rate point, learned solutions considerably outperform previous fingerprint coding standards, like JPEG2000, both in terms of distortion and minutiae preservation. Indeed, experimental results prove that the peculiarities of learned compression artifacts do not prevent automatic fingerprint identification (since minutiae types and locations are not significantly altered), nor do compromise image quality for human visual inspection (as they gain in terms of BD rate and PSNR of 47.8% and +3.97dB respectively).
Paper Structure (12 sections, 3 equations, 4 figures)

This paper contains 12 sections, 3 equations, 4 figures.

Table of Contents

  1. Introduction
  2. Background
  3. Fingerprints
  4. Image and Fingerprint compression
  5. Method
  6. Mean and Scale Hyperprior
  7. Finger-MSH
  8. Experiments and Results
  9. Dataset and training
  10. Evaluation metrics
  11. Results
  12. Conclusions

Figures (4)

  • Figure 1: Learned image codec with a mean and scale hyperprior entropy model used as the main architecture in this work.
  • Figure 2: Original at $\sim$ 74kB (top-left), compressed with Finger-MSH at $\sim$ 4kB, PSNR=24.84dB (top-right), compressed with MSH at $\sim$ 5kB, PSNR=23.03dB (bottom-left), compressed with JPEG2000 at $\sim$ 5kB, PSNR=17.79 dB (bottom-right)
  • Figure 3: RD curves for PSNR (top) and SSIM (bottom)
  • Figure 4: Correctly kept minutiae (left column), extra introduced minutiae (central column), and minutiae that changed type (right column) as a function of rate (first row) and SSIM (second row).