Fingerprint Identification using Modified Capsule Network

With the ever-increasing usage of biometric systems in today's scenario, there's a need for accurate identification of fingerprints. The fingerprint data is associated with various important services in today's scenario like aadhar verification, mobile unlocking, biometric attendance and the security aspects of the devices and accounts. To utilize the fingerprint data, it needs to be analyzed efficiently. The fingerprint identification process needs to look at various challenges like even if the fingerprint is rotated, altered, or not completely available still a correct prediction is required. Many deep learning algorithms like Convolution Neural Network (CNN), Inception V3 and Capsule Network have been implemented in this segment still there is a need to design the algorithm for the same with higher accuracy. In this work, a Modified Capsule Network is proposed for effective fingerprint identification. The experiments were performed utilizing the biometric Sokoto Coventry Fingerprint (SOCOFing) dataset extracted from Kaggle. The proposed model achieves better accuracy than the state-of-the-art models in this segments. It approaches more than 99 % accuracy for classifying fingerprints into fingers, hands, and gender class category.