Identification of Distorted Fingerprints Using Wavelet Method and Convolutional Neural Network (CNN)

Abstract

Biometrics offers a valuable tool for disaster victim identification, particularly through fingerprints. However, distorted or damaged fingerprints pose a significant challenge for recognition. This study explores the potential of Wavelet and Convolutional Neural Network (CNN) techniques to enhance the accuracy of distorted fingerprint recognition. Wavelet transform addresses the non-stationary nature of images and reduces detected noise. Convolutional Autoencoder, a CNN component, generates simplified feature representations from input images and attempts to reconstruct them. Utilizing 500 fingerprint samples, the testing results demonstrate accuracy variations ranging from 11% to 59.2%. Image reconstruction achieved 7.16% to 12.47% accuracy, while fingerprint matching attained accuracies between 92.71% and 93.96%. Averaging across all damage levels, the overall accuracy reached 37.65%, with average fingerprint reconstruction at 9.31% and average matching accuracy at 93.03%. The successful reconstruction and matching of distorted fingerprints within a certain range of damage using Wavelet and Convolutional Neural Network highlights the promising potential of these techniques for improved fingerprint identification in forensic and security contexts.