Scholarlink

<p>Fingerprint recognition is a widely used biometric technique, yet its accuracy is often compromised by noise and non-uniform contact during image capture. To address these challenges, a novel structure-preserving Generative Adversarial Network (GAN) embedded with Singular Value Decomposition (SVD), termed SVD-Net, is proposed for fingerprint image enhancement. The SVD-Net, built on a GAN network architecture with integrated SVD processing layers, leverages the inherent structural properties of fingerprints through SVD to enhance denoising performance while preserving critical ridge and valley details. The model simultaneously performs denoising and inpainting, effectively handling weak signals and background clutter. Experimental results demonstrate that SVD-Net outperforms comparison methods, achieving superior performance on key evaluation metrics, including MSE (0.0275), PSNR (15.73), and SSIM (0.7955). The proposed approach not only enhances fingerprint image quality but also introduces a novel network design perspective, providing a robust framework for fingerprint enhancement.</p>

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

SVD-Net: structure-preserving GAN with embedded SVD for fingerprint image enhancement

  • 期刊论文

  • 发表日期：

    2025年9月11日

查看全文
  • Wei Zhong,
  • Ruiwen Li

摘要

Fingerprint recognition is a widely used biometric technique, yet its accuracy is often compromised by noise and non-uniform contact during image capture. To address these challenges, a novel structure-preserving Generative Adversarial Network (GAN) embedded with Singular Value Decomposition (SVD), termed SVD-Net, is proposed for fingerprint image enhancement. The SVD-Net, built on a GAN network architecture with integrated SVD processing layers, leverages the inherent structural properties of fingerprints through SVD to enhance denoising performance while preserving critical ridge and valley details. The model simultaneously performs denoising and inpainting, effectively handling weak signals and background clutter. Experimental results demonstrate that SVD-Net outperforms comparison methods, achieving superior performance on key evaluation metrics, including MSE (0.0275), PSNR (15.73), and SSIM (0.7955). The proposed approach not only enhances fingerprint image quality but also introduces a novel network design perspective, providing a robust framework for fingerprint enhancement.