A Method based on Convolutional Neural Networks for Fingerprint Segmentation

In forensic science, the resolution of crimes is associated with the identification of those involved. In the civil context, the security of automated processes depends on the identification of authorized people. In this sense, fingerprint-based recognition techniques stand out. A fundamental stage is the calculation of the degree of similarity between the samples presented, so the task of identifying a region of interest (ROI), excluding noisy regions, can improve the precision and reduce the computational cost. In this aspect, this work presents a technique of segmentation of the region of interest based on convolutional neural networks (CNN) without pre-processing steps. The new approach was evaluated in two different architectures from state of the art, presenting similarity indexes Distance of Hausdorff (5.92), Dice coefficient (97.28%) and Jaccard Similarity (96.77%) superior to the classic methods. The error rate (3.22%) was better than five segmentation techniques from state of the art and showed better results than another deep learning approach, presenting promising results to identify the region of interest with potential for application in systems based on biometric identification.