Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Generative Iris Prior Embedded Transformer for Iris Restoration

Yubo Huang, Jia Wang, Peipei Li, Liuyu Xiang, Peigang Li, Zhaofeng He

TL;DR

Iris restoration under complex degradation is addressed by Gformer, a hierarchical encoder–decoder that embeds a pretrained generative iris prior into a Transformer-based restoration pipeline. The model uses depth-wise self-attention for efficiency and an Iris Feature Modulator to fuse multi-resolution features with StyleGANv2 priors, guided by pyramid and perceptual losses to enhance realism and recognition fidelity. Experiments on CASIA-Iris-Lamp and CASIA-Iris-Thousand show state-of-the-art recognition gains, validating the approach despite not always achieving the highest pixel-level scores. The work demonstrates that priors from generative models can substantially improve recognition performance in degraded iris imagery, enabling more robust, large-scale deployment.

Abstract

Iris restoration from complexly degraded iris images, aiming to improve iris recognition performance, is a challenging problem. Due to the complex degradation, directly training a convolutional neural network (CNN) without prior cannot yield satisfactory results. In this work, we propose a generative iris prior embedded Transformer model (Gformer), in which we build a hierarchical encoder-decoder network employing Transformer block and generative iris prior. First, we tame Transformer blocks to model long-range dependencies in target images. Second, we pretrain an iris generative adversarial network (GAN) to obtain the rich iris prior, and incorporate it into the iris restoration process with our iris feature modulator. Our experiments demonstrate that the proposed Gformer outperforms state-of-the-art methods. Besides, iris recognition performance has been significantly improved after applying Gformer.

Generative Iris Prior Embedded Transformer for Iris Restoration

TL;DR

Iris restoration under complex degradation is addressed by Gformer, a hierarchical encoder–decoder that embeds a pretrained generative iris prior into a Transformer-based restoration pipeline. The model uses depth-wise self-attention for efficiency and an Iris Feature Modulator to fuse multi-resolution features with StyleGANv2 priors, guided by pyramid and perceptual losses to enhance realism and recognition fidelity. Experiments on CASIA-Iris-Lamp and CASIA-Iris-Thousand show state-of-the-art recognition gains, validating the approach despite not always achieving the highest pixel-level scores. The work demonstrates that priors from generative models can substantially improve recognition performance in degraded iris imagery, enabling more robust, large-scale deployment.

Abstract

Iris restoration from complexly degraded iris images, aiming to improve iris recognition performance, is a challenging problem. Due to the complex degradation, directly training a convolutional neural network (CNN) without prior cannot yield satisfactory results. In this work, we propose a generative iris prior embedded Transformer model (Gformer), in which we build a hierarchical encoder-decoder network employing Transformer block and generative iris prior. First, we tame Transformer blocks to model long-range dependencies in target images. Second, we pretrain an iris generative adversarial network (GAN) to obtain the rich iris prior, and incorporate it into the iris restoration process with our iris feature modulator. Our experiments demonstrate that the proposed Gformer outperforms state-of-the-art methods. Besides, iris recognition performance has been significantly improved after applying Gformer.
Paper Structure (12 sections, 11 equations, 3 figures, 2 tables)

This paper contains 12 sections, 11 equations, 3 figures, 2 tables.

Table of Contents

  1. Introduction
  2. Proposed Method
  3. Overview
  4. Depth-Wise Self-Attention
  5. Iris Feature Modulator
  6. Model Objectives
  7. Experiments
  8. Datasets
  9. Implementation Details
  10. Comparisons with State-of-the-Art Methods
  11. Ablation Study
  12. Conclusion

Figures (3)

  • Figure 1: Overview of Gformer architecture. Gformer consists of Transformer encoder and generative iris prior embedded decoder. They are bridged by the iris feature modulator. (a) shows the core component of Transformer block: depth-wise self-attention that operates attention score across channels rather than spatial dimension. (b) shows the spatial feature transform in the iris feature modulator, which leverages generative iris prior.
  • Figure 2: Qualitative comparison with state-of-the-art methods.
  • Figure 3: ROC curve for comparison with state-of-the-art methods.