Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Metric for Evaluating Performance of Reference-Free Demorphing Methods

Nitish Shukla, Arun Ross

TL;DR

Addresses the challenge of evaluating reference-free demorphing where recovering constituent faces from a morph is ill-posed. Proposes biometrically cross-weighted IQA (BW-IQA), which combines image-quality metrics with biometric similarity to jointly assess identity preservation and visual fidelity. Benchmarks three demorphing methods (IPD, SDeMorph, Facial Demorphing) across six datasets using two matchers (AdaFace, ArcFace), showing BW-IQA provides consistent rankings and avoids the misleading results produced by TMR, RA, or pure IQA metrics. The work delivers a practical, generalizable evaluation framework with clear implications for security-sensitive deployment of demorphing techniques.

Abstract

A facial morph is an image created by combining two (or more) face images pertaining to two (or more) distinct identities. Reference-free face demorphing inverts the process and tries to recover the face images constituting a facial morph without using any other information. However, there is no consensus on the evaluation metrics to be used to evaluate and compare such demorphing techniques. In this paper, we first analyze the shortcomings of the demorphing metrics currently used in the literature. We then propose a new metric called biometrically cross-weighted IQA that overcomes these issues and extensively benchmark current methods on the proposed metric to show its efficacy. Experiments on three existing demorphing methods and six datasets on two commonly used face matchers validate the efficacy of our proposed metric.

Metric for Evaluating Performance of Reference-Free Demorphing Methods

TL;DR

Addresses the challenge of evaluating reference-free demorphing where recovering constituent faces from a morph is ill-posed. Proposes biometrically cross-weighted IQA (BW-IQA), which combines image-quality metrics with biometric similarity to jointly assess identity preservation and visual fidelity. Benchmarks three demorphing methods (IPD, SDeMorph, Facial Demorphing) across six datasets using two matchers (AdaFace, ArcFace), showing BW-IQA provides consistent rankings and avoids the misleading results produced by TMR, RA, or pure IQA metrics. The work delivers a practical, generalizable evaluation framework with clear implications for security-sensitive deployment of demorphing techniques.

Abstract

A facial morph is an image created by combining two (or more) face images pertaining to two (or more) distinct identities. Reference-free face demorphing inverts the process and tries to recover the face images constituting a facial morph without using any other information. However, there is no consensus on the evaluation metrics to be used to evaluate and compare such demorphing techniques. In this paper, we first analyze the shortcomings of the demorphing metrics currently used in the literature. We then propose a new metric called biometrically cross-weighted IQA that overcomes these issues and extensively benchmark current methods on the proposed metric to show its efficacy. Experiments on three existing demorphing methods and six datasets on two commonly used face matchers validate the efficacy of our proposed metric.
Paper Structure (10 sections, 9 equations, 2 figures, 3 tables)

This paper contains 10 sections, 9 equations, 2 figures, 3 tables.

Table of Contents

  1. Introduction
  2. Face Demorphing
  3. Previous Work
  4. Evaluation Metrics
  5. Evaluation Criterion
  6. Biometrically cross-weighted IQA
  7. Datasets
  8. Experiments
  9. Results
  10. Conclusion

Figures (2)

  • Figure 1: Single-image reference-free demorphing: The MORPH image is created by blending ground-truth (GT) face images. Reconstructions are produced using Identity Preserving Decomposition (IPD) ref66, SDeMorph ref18, and Facial Demorphing ref51. Among these, IPD achieves the best visual quality, which aligns with the score generated by our proposed metric.
  • Figure 2: Comparison of the efficacy of SSIM and PSNR with that of the proposed metric for demorphing ($\epsilon=0.3$). (Top) The middle image is more structurally similar to the blurred image on right compared to the noisy image of the same subject on the left. (Bottom) $\mathcal{I}_1$ is more similar to $\mathcal{I}_2$, which belong to different subjects, compared to the noisy version of $\mathcal{I}_1$. Our proposed metric correctly balances the identity and image quality to produce consistent scores.