Research And Application Of Cancelable Palmprint Template Protection Algorithm Based On Multiple Protection Mechanisms

With the development of information technology,traditional identity authentication methods based on password tokens no longer meet user needs.Biometric recognition systems are widely used for identity authentication due to their high recognition accuracy and efficiency advantages.The recognition of biometric systems mainly relies on human physiological features for recognition.Due to the uniqueness and long-term invariance of user physiological features,it will lead to severe information leakage issues once attacked.In addition,a performance analysis was conducted on the TJU palmprint dataset.Convolutional neural networks can not only efficiently process a large number of palmprint images but also extract deeper palmprint feature information through continuous learning.Therefore,we propose a cancelable palmprint template protection scheme based on deep hashing and attention mechanism,which protects palm print templates through an end-toend convolutional network model.The main improvements to the model include(1)fusing spatial and channel attention based on Res Net50 to improve the performance of model.(2)Using chaotic mapping to design revocable layers,different palm print feature templates are generated by changing the initial value of the chaotic mapping to control the deactivation of some neurons.(3)To address the high memory consumption caused by the high dimensionality of palmprint depth features,a hash layer is designed in the model to reduce feature dimensions and improve retrieval efficiency.A cancelable palmprint template protection scheme based on a random sampling mechanism and relocation bloom filter is proposed to address the issues of low recognition performance and hash conflicts in existing template protection schemes based on the bloom filter.Firstly,a key management system was designed to generate keys and chaotic matrices.Secondly,the random sampling mechanism to generate a security matrix.Finally,the secure matrix is mapped into a cancelable palmprint feature template using a parallel structure relocation bloom filter.In addition,to explore the scalability of solution,we conducted performance analysis on the FVC 2002(DB1,DB2,DB3)and FVC 2004(DB1,DB2,DB3)fingerprint datasets.To address the security issues of palmprint recognition systems caused by external factor leakage,a one-factor cancelable palm template protection scheme is proposed based on the previous research,which protects palmprint features through multiple protection mechanisms.Firstly,a network utilizing deep hashing and attention mechanisms extracts palmprint feature values.Secondly,the extracted palmprint feature values are randomly sampled to generate a more informative security matrix.Finally,generate a cancelable palmprint template through the relocation bloom filter.In addition,performance analysis experiments were conducted on the TJU palmprint dataset.