Fingerprint and Palm Print Recognition Systems based on Recurrent Neural Network

In recent years, biometric authentication has become essential in fields of security and access control systems. However, traditional Convolutional Neural Network (CNN)-based biometric systems encounter some challenges, including limited ability to modeling spatiotemporal dependencies in fingerprint and palmprint images. To overcome these challenges, this research proposes a Recurrent Neural Network (RNN) model to classify Local Binary Pattern (LBP) feature maps by learning spatial and temporal representations. Initially, data is gathered from CASIA-FingerprintV5 for fingerprint recognition and CASIA-PalmprintV1 for palmprint recognition, which contained fingerprint as well as palm print images. These images are further preprocessed with Histogram Equalization Technique (HET) to enhance contrast of input images by adjusting pixel values to create uniform distribution. Also, making it easier to distinguish between different patterns and features in the fingerprint or palm print. The normalized images are converted to LBP feature maps with LBP technique, then fed into RNN model. This model classifies LBP feature maps by learning spatiotemporal dependences with recurrent connections (loops), enabling it to recognize fingerprints and palmprints. Finally, the proposed RNN outperformed the existing CNN by achieving better results in terms of recognition accuracy (99.87%).