目的 近年来，指纹识别技术被大规模应用于我们的日常生活中，如身份鉴定、指纹支付与考勤等。然而，最新研究表明这些系统极易遭受伪造指纹的欺骗攻击，因此在使用指纹认证用户身份前，鉴别待测指纹的真伪至关重要。伪造指纹的制作材料具有多样性，现有工作忽视了不同材料伪造指纹之间数据分布的关联性，致使跨材料检测泛化性普遍较低。因此，本文通过分析不同材料伪造指纹数据间的分布关联性，挖掘不同伪造指纹间的材料域不变伪造特征，提出了一种基于共性特征学习的高泛化伪造指纹检测方法。方法 首先，为了表征和学习不同材料伪造指纹间的特征，设计了一种多尺度伪造特征提取器(MSFE, Multi-scale Spoofing Feature Extractor)，包含一个多尺度空间通道(MSC, Multi-scale Spatial-Channel)注意力模块，以学习真假指纹类间的细粒度差异特征。然后，为了进一步分析不同材料伪造指纹数据间的分布关联性，又构造了一种共性伪造特征提取器(CSFE, Common Spoofing Feature Extractor)，在MSFE先验知识的引导下进行多任务的材料域不变伪造特征学习。最后，设计一个材料鉴别器对学习到的共性伪造特征进行约束，同时构建一个自适应联合优化损失模块来平衡多个模块在训练过程中的损失权重，以进一步提高面对未知材料伪造指纹检测时的泛化性。结果 在两个公开的指纹数据集(LivDet2017和LivDet2019)上进行了跨材料测试，实验结果表明所提算法相比于现有工作，ACE降低了1.34%，TDR提高了1.43%，展现出较高的泛化性。结论 本文所提的算法相较于现有方法，在ACE和TDR方面均取得最佳的性能。此外，当面对未知材料的伪造指纹检测时，同样展现出最优的泛化性。

Abstract: Objective In recent times, the realm of our daily lives has witnessed the ubiquitous integration of fingerprint recognition technology, permeating domains such as authorizing identification, fingerprint-based payments, and access control systems. However, recent studies have shown that these systems are vulnerable to spoofing fingerprint attacks. Attackers can deceive authentication systems by imitating fingerprints with artificial materials. Thus, an imperative arises to ascertain the authenticity of the fingerprint under scrutiny prior to employing it as a means to authenticate the user"s identity. The development of spoofing fingerprint detection technology has attracted extensive attention from academia and industry. The materials employed in the creation of spoofing fingerprints exhibit diversity. The present research disregards the correlation in data distribution among spoofing fingerprints crafted from various materials, consequently leading to limited generalization in cross-material detection. Hence, through analyzing the distribution correlation among counterfeit fingerprint data originating from diverse materials, and the exploration of invariant forgery features within the material domain of distinct counterfeit fingerprints, this paper proposes a high generalization spoofing fingerprint detection method based on commonality feature learning. Method Firstly, in order to characterize and learn the features between spoofing fingerprints of different materials, a Multi-scale Spoofing Feature Extractor (MSFE) is designed, which includes a Multi-scale Spatial-Channel (MSC) attention module, so that multi-scale spoofing feature extractor pays more attention to the fine-grained differences between live and fake fingerprints and improves the ability of the network to learn spoofing features. Then, in order to further analyze the distribution correlation between spoofing fingerprint data of different materials and extract the common spoofing features between spoofing fingerprints of various materials, a Common Spoofing Feature Extractor (CSFE) is constructed. Under the guidance of multi-scale spoofing feature extractor prior knowledge, common spoofing feature extractor calculates the distance of the feature distribution extracted by multi-scale spoofing feature extractor and common spoofing feature extractor in the regenerated Hilbert space through the feature distance measurement module, and minimizes the maximum mean difference (MMD) of the data distribution to reduce the distance between the two. The multi-task material domain invariant spoofing feature learning is carried out, and a material discriminator is designed to constrain the learned common spoofing features and remove the specific material information of the spoofing fingerprint. Common spoofing feature extractor involves the calculation of multiple loss functions. Manually setting the weight ratio of these loss functions may cause the model performance to be unable to improve. Therefore, we use an adaptive joint optimization loss function to balance the loss values of each module, and further improve the generalization ability of the network in the face of unknown material spoofing fingerprints. In the training process, the fingerprint image used contains two kinds of labels, which are the authenticity label of the fingerprint and the material label of the forged fingerprint. The true fingerprint has no material properties and is marked as 0. Forged fingerprints are numbered from 1 according to the material category, the authenticity of fingerprints is judged according to the authenticity label, and the type of forged materials is judged according to the material label. The random gradient descent method is used for optimization, and the learning rate setting is trained from 0.001, which is reduced by 0.1 times per 10 epochs. Result The experimental results on two public datasets show that the algorithm proposed in this paper achieves the best comprehensive performance in the cross-material detection of forged fingerprints. On the GreenBit sensor of LivDet2017 dataset, ACE reduces the rate by 0.16% compared with the second-ranked spoofing fingerprint detection model and increases TDR by 2.4%; On the Digitalpersona sensor of LivDet2017 dataset, ACE reduces the rate by 0.26% compared with the second-ranked forgery fingerprint detection model and increases TDR by 0.7%; On LivDet2019 dataset, ACE reduces the rate by 1.34% on average compared with the second-ranked spoofing fingerprint detection model and increases TDR by 1.43% on average, showing higher generalization. In order to verify the superiority of the MSC attention module compared with the CBAM module for spoofing fingerprint detection, a comparative experiment was also performed. To better evaluate our method, we also conducted a series of ablation experiments to verify each module involved in common feature extraction training contributes to the cross-material spoofing fingerprint detection task. In order to show the improvement of generalization performance of common spoofing feature extractor compared with multi-scale spoofing feature extractors in cross-material spoofing fingerprint detection, this paper visualizes the distribution of the proposed features by t-SNE algorithm. Conclusion The method proposed in this paper achieves better detection results than other methods and shows higher generalization performance when detecting spoofing fingerprints made of unknown materials. Compared with the spoofing fingerprint detection under the same material, the extant spoofing fingerprint detection technique yet harbors substantial scope for refinement in terms of its generalization capabilities for cross-material detection. The task of cross-material spoofing fingerprint detection more aptly aligns with practical requirements and bears immense significance within the realm of research pursuits.