New approach for fingerprint recognition based on stylometric features with blockchain and cancellable biometric aspects

【Author】 Elsadai, Ali; Adamovic, Sasa; Sarac, Marko; Saracevic, Muzafer; Kumar Sharma, Sudhir

【Source】MULTIMEDIA TOOLS AND APPLICATIONS

【影响因子】2.577

【Abstract】Applying machine learning techniques and methods in biometric recognition has gained significant attention in recent years as it can provide a better performance, high accuracy, and cancellable biometrics data. This paper proposes a new approach for fingerprint recognition based on machine learning methods and stylometric features. The proposed solution deals with fingerprint recognition, cancellability, stylometry, blockchain and machine learning. This research uses machine learning methods that classify fingerprint templates as a numeric feature instead of using Gabor wavelets and filters. The proposed method gives very high accuracy for biometric fingerprint templates. For these reasons, we additionally consider the use of an internal blockchain in the form of a distributed database that implements all security services, including privacy protection. Because the recognition method is based on machine learning, the generated templates are a numerical data type and take up minimal memory size, which further favors the application of a blockchain and enables implementation even in IoT devices. We generate the fingerprint biometric template by converting an enhanced fingerprint image into a 1-D set of fixed length codes. After that, we extract stylometric features that will be used for classification. The experiment is conducted on the CASIA-FingerprintV5 and achieved excellent results where the CatBoost method with over-sampling (SMOTE) achieved the best results for All_features(42) and GRRF(10) sets with 99.95% accuracy and 99.98%, respectively, and FAR 0.0007 and 0.0003, respectively. In addition, the proposed system significantly decreased the computational costs which makes it suitable for other applications.

【Keywords】Machine learning; Ensemble learning; Random forest; Feature selection; Blockchain

【发表时间】

【收录时间】2022-07-25

【文献类型】理论性文章

【主题类别】

区块链应用-实体经济-其他