FDWST: Fingerphoto Deblurring using Wavelet Style Transfer

TL;DR

FDWST introduces a fingerphoto deblurring architecture that fuses 2D discrete wavelet transforms with a wavelet-domain style transfer to restore sharpness while preserving fingerprint minutiae. The method replaces the encoder with a Level-3 DWT, applies a per-band style transfer using a binarized sharp source, and reconstructs the image via a decoder, guided by content, style, and identity losses. Across training on the WVU fingerphoto dataset and evaluation with VeriFinger and IDKit, FDWST achieves state-of-the-art matching performance (AUC up to ~0.99, EER ~0.05) and notable improvements in perceptual quality. The work demonstrates that wavelet-domain sharpness transfer can effectively recover high-frequency details without sacrificing identity, suggesting broad applicability to practical biometric deblurring scenarios.

Abstract

The challenge of deblurring fingerphoto images, or generating a sharp fingerphoto from a given blurry one, is a significant problem in the realm of computer vision. To address this problem, we propose a fingerphoto deblurring architecture referred to as Fingerphoto Deblurring using Wavelet Style Transfer (FDWST), which aims to utilize the information transmission of Style Transfer techniques to deblur fingerphotos. Additionally, we incorporate the Discrete Wavelet Transform (DWT) for its ability to split images into different frequency bands. By combining these two techniques, we can perform Style Transfer over a wide array of wavelet frequency bands, thereby increasing the quality and variety of sharpness information transferred from sharp to blurry images. Using this technique, our model was able to drastically increase the quality of the generated fingerphotos compared to their originals, and achieve a peak matching accuracy of 0.9907 when tasked with matching a deblurred fingerphoto to its sharp counterpart, outperforming multiple other state-of-the-art deblurring and style transfer techniques.

Figures (11)

• Figure 1: Sample pairs of sharp and blurry fingerphotos, with the deblurred version in the middle column.
• Figure 2: A sharp fingerphoto alongside its corresponding level 1 wavelet decomposition. The top left sub-band is an approximation of the original image, being a smaller smoothed version. The remaining three sub-bands contain horizontal, vertical, and diagonal information, respectively.
• Figure 3: Our proposed Wavelet Style Transfer (WST) module, which enhances a blurry set of wavelets $W_b^{(i)}$ with a sharp set $W_s^{(i)}$, and produces a style transferred set $W_t^{(i)}$.
• Figure 4: Sample IDWT results from a Level 3 DWT. The Blurry photo on the left is fixed, and we perform WST with three different sharp images before inverting. The first sharp image is the same ID fingerphoto as the blurry one, the second is a different ID, and the third is a binarized different ID.
• Figure 5: Examples of IDWT reconstructions from different level DWTs. The lower the level, the blurrier the reconstruction. With higher levels, artifacts become prevalent and the identities mix. Finally, at Level 8, the sharp identity fully replaces the original blurry one.