Regression of Dense Distortion Field from a Single Fingerprint Image

TL;DR

This work tackles the problem of skin distortion degrading fingerprint matching by moving beyond PCA-based distortion models to directly regressing a dense distortion field from a single distorted fingerprint. It introduces an end-to-end self-reference network with texture and orientation branches and a multi-scale fusion module to predict a dense $16\times16$ displacement field for accurate rectification, independent of finger pose. The method is trained on a large synthetic dataset (TDF_V2) generated by augmenting real distortion patterns, and it achieves state-of-the-art distortion estimation and improved matching across multiple databases and matchers, including minutiae-based and fixed-length descriptor methods. The practical impact is more reliable fingerprint authentication in diverse distortion scenarios, with robust rectification enabling better downstream matching in real-world applications.

Abstract

Skin distortion is a long standing challenge in fingerprint matching, which causes false non-matches. Previous studies have shown that the recognition rate can be improved by estimating the distortion field from a distorted fingerprint and then rectifying it into a normal fingerprint. However, existing rectification methods are based on principal component representation of distortion fields, which is not accurate and are very sensitive to finger pose. In this paper, we propose a rectification method where a self-reference based network is utilized to directly estimate the dense distortion field of distorted fingerprint instead of its low dimensional representation. This method can output accurate distortion fields of distorted fingerprints with various finger poses and distortion patterns. We conducted experiments on FVC2004 DB1\_A, expanded Tsinghua Distorted Fingerprint database (with additional distorted fingerprints in diverse finger poses and distortion patterns) and a latent fingerprint database. Experimental results demonstrate that our proposed method achieves the state-of-the-art rectification performance in terms of distortion field estimation and rectified fingerprint matching.

Figures (14)

• Figure 1: Impact of the proposed rectification algorithm on fingerprint matching scores. The red and blue numbers over arrows represent the matching scores before and after rectification calculated by VeriFinger SDK 12.0 VeriFinger. Both distorted-undistorted and distorted-distorted matching scores are significantly improved after rectification (105 $\rightarrow$ 279, 183 $\rightarrow$ 341, 46 $\rightarrow$ 208).
• Figure 2: Failure cases of PCA based rectification. (a) Distortion details are lost. Green and orange rectangles represent regions where orientation or magnitude information is lost. (b) Examples where finger pose is estimated inaccurately. The red arrow indicates the finger pose extimated by yin2021joint.
• Figure 3: Flowchart of the proposed distorted fingerprint rectification system.
• Figure 4: Structure of the proposed distortion estimation network. The network includes a texture branch and an orientation branch for feature extraction, a spatial pyramid block for fusing multi-scale feature information, and a convolution regression block for predicting dense distortion field. For an input distorted fingerprint, the network takes the $16 \times 16$ pixel block as a unit and gives its corresponding distortion field (displacement from the input image to the rectification target).
• Figure 5: Examples of fingerprint preprocessing. (a) and (b) are respectively collected from the optical sensor and cirme scene. A and B represent different preprocessing strategies.