Generating Realistic Forehead-Creases for User Verification via Conditioned Piecewise Polynomial Curves

TL;DR

The paper tackles the data scarcity and cross-domain generalization challenge in forehead-crease biometrics by introducing a geometry-driven synthesis pipeline that models crease patterns with $B$-splines and Bézier curves on a dynamic $6 \times 6$ grid. It trains a diffusion-based Image-to-Image translator (Edge2FC) to convert geometrically generated edge maps into realistic crease appearances, and amplifies intra-subject diversity via Control Point Diversity (CPD) and Visual Prompt Diversity (VPD). A training curriculum progressively blends synthetic and real data, enabling robust cross-database verification with the FHCVS backbone, and achieving measurable gains in EER and TMR over baselines. Despite limitations in texture fidelity, the approach demonstrates that geometry-driven prompts, paired with diffusion translation and curriculum-based learning, can meaningfully enhance forehead-crease verification in challenging real-world scenarios.

Abstract

We propose a trait-specific image generation method that models forehead creases geometrically using B-spline and Bézier curves. This approach ensures the realistic generation of both principal creases and non-prominent crease patterns, effectively constructing detailed and authentic forehead-crease images. These geometrically rendered images serve as visual prompts for a diffusion-based Edge-to-Image translation model, which generates corresponding mated samples. The resulting novel synthetic identities are then used to train a forehead-crease verification network. To enhance intra-subject diversity in the generated samples, we employ two strategies: (a) perturbing the control points of B-splines under defined constraints to maintain label consistency, and (b) applying image-level augmentations to the geometric visual prompts, such as dropout and elastic transformations, specifically tailored to crease patterns. By integrating the proposed synthetic dataset with real-world data, our method significantly improves the performance of forehead-crease verification systems under a cross-database verification protocol.

Figures (7)

• Figure 1: Proposed Method: Geometrically rendered visual prompts use an Edge-to-Forehead Creases model to generate novel synthetic IDs. Principal creases are rendered using B-spine curves on $j^{th}$ row $r^c_j$ of a $6 \times 6$ grid with start, control and end points: $s^c_j$, $p^c_j$, and, $e^c_j$, while non-prominent creases are plotted using Bézier curves on $r^w_j$ formed by merging two consecutive cells $u_k^i$. The spatial position for each crease is decided by a random grid mask $M$. The rounded box with a light-blue background (top) indicates the training phase, while the remaining three in a medium-purple background (bottom) denotes the inference stage pipeline. Best viewed in color.
• Figure 2: Edge Extraction: A forehead-creases image at each pre-processing stage (a)-(f) to obtain the final edge map (f).
• Figure 3: Comparison: Synthetic IDs from (a) SA-PermuteAug tandon2024synthetic, (b) BSpline-VPD, and, (c) DiffEdges-VPD datasets. DiffEdges-VPD offers higher realism (Table \ref{['table:fid']}), however both offer comparable quantitative performance (Table \ref{['table:eer']}).
• Figure 4: Intermediate feature maps visualization for images from (a) FH-V1 (Real), (b) BSpline-VPD, and, (c) DiffEdges-VPD datasets. Despite high FID scores, the verification model mainly focuses on creases rendered using b-splines in (b), while both creases and textures in (a) and (c). Features maps resized from $(112, 112)$ for a better visualization.
• Figure 5: Combined DET curve for comparative analysis among all different experiments.The x-axis is in log scale.