[论文解读] ResWCAE: Biometric Pattern Image Denoising Using Residual Wavelet-Conditioned Autoencoder

Youzhi Liang, Liang Wen|arXiv (Cornell University)|Jul 23, 2023

Biometric Identification and Security被引用 46

一句话总结

ResWCAE 引入一种轻量级残差小波条件卷积自编码器，并采用 KL 散度正则化来降噪指纹图像，在强噪声下对比最先进方法具有更好性能。

ABSTRACT

The utilization of biometric authentication with pattern images is increasingly popular in compact Internet of Things (IoT) devices. However, the reliability of such systems can be compromised by image quality issues, particularly in the presence of high levels of noise. While state-of-the-art deep learning algorithms designed for generic image denoising have shown promise, their large number of parameters and lack of optimization for unique biometric pattern retrieval make them unsuitable for these devices and scenarios. In response to these challenges, this paper proposes a lightweight and robust deep learning architecture, the Residual Wavelet-Conditioned Convolutional Autoencoder (Res-WCAE) with a Kullback-Leibler divergence (KLD) regularization, designed specifically for fingerprint image denoising. Res-WCAE comprises two encoders - an image encoder and a wavelet encoder - and one decoder. Residual connections between the image encoder and decoder are leveraged to preserve fine-grained spatial features, where the bottleneck layer conditioned on the compressed representation of features obtained from the wavelet encoder using approximation and detail subimages in the wavelet-transform domain. The effectiveness of Res-WCAE is evaluated against several state-of-the-art denoising methods, and the experimental results demonstrate that Res-WCAE outperforms these methods, particularly for heavily degraded fingerprint images in the presence of high levels of noise. Overall, Res-WCAE shows promise as a solution to the challenges faced by biometric authentication systems in compact IoT devices.

研究动机与目标

• 在高噪声水平下为紧凑型物联网设备提供可靠的生物特征去噪的动机。
• 开发面向指纹模式而非通用图像的轻量级架构。
• 通过残差连接和小波域条件化来保留指纹的细粒度空间特征。
• 引入 KL-散度正则化以提高泛化能力并防止过拟合。

提出的方法

• 提出带有两个编码器（图像编码器和小波编码器）和一个解码器的 Res-WCAE。
• 在编码器和解码器之间使用残差连接以保留空间细节。
• 将解码器条件化为来自图像和小波编码器的融合表示；通过对小波子图应用三层卷积神经网络来包含小波域特征。
• 在损失中应用 KL-散度正则化：L(Θ)=E[I]||y_D^[L]-I||^2 + λ E[y] D_KL(y_D^[L] || I)。
• 将输入处理为二维灰度指纹；利用 Symlets 小波分解最高到三层作为小波输入。
• 在 SOCOFing 数据集上对比基线模型进行训练和评估，AWGN 噪声水平 σ 在 [100,200]。

Figure 1: A schematic for the architecture of a Residual Wavelet-Conditioned Convolutional Autoencoder (ResWCAE), including a sample noise-free image, ${\bm{I}(103,96)}$ , and a denoised image, $\hat{\bm{I}}(103,96)$ .

实验结果

研究问题

• RQ1 在强噪声条件下，Res-WCAE 是否能超越现有的指纹去噪方法？
• RQ2 在解码器上同时对图像域和小波域特征进行条件化是否能改善指纹细节（纹理点）的恢复？
• RQ3 在紧凑设备上的指纹去噪中，KL-散度正则化是否有利于泛化？
• RQ4 在不同噪声条件下，Res-WCAE 相对于最先进模型（如 U-Finger、Fpd-m-net）的表现如何？

主要发现

模型	σ=100	PSNR	SSIM	MSE
AWGN	7.92	0.45	0.17
Dense NN	10.03	0.35	0.12
Autoencoder	12.77	0.69	0.06
Res-WCAE	17.88	0.79	0.02

• 在 AWGN-指纹去噪中，当 σ=100 时，Res-WCAE 在所比模型中取得了最高的 PSNR 和 SSIM，且 MSE 最低。
• Res-WCAE 的结果为 PSNR=17.88, SSIM=0.79, MSE=0.02，优于 Autoencoder (PSNR=12.77, SSIM=0.69, MSE=0.06) 和 Dense NN (PSNR=10.03, SSIM=0.35, MSE=0.12)。
• Res-WCAE 在所报道的评估中优于最先进的指纹去噪模型 U-Finger 和 Fpd-m-net。
• 该方法在更广的噪声范围内（σ 高达 200）表现出鲁棒性，能够在去噪后的指纹中保留细微纹理。
• 在广泛的噪声水平范围内平均提升约 7.5 dB 的 PSNR。

Figure 2: Samples of original figures, noisy figures and denoised figures for noise level $\sigma$ from 100 to 200.. Model A: placeholder, Model B: placeholder, Model C: placeholder, Model D: Res-WCAE.