Deep Learning Method for Multi-Attribute Analysis of Fingerprint Images

Abstract

Estimation of gender, hand, and finger to minimize the probable suspects list in a fingerprint database search is a very important stride in forensic anthropology. Previous research attempted to estimate the gender, hand, and finger from the fingerprint, but the results were not consistent. In this effort, we proposed gender, hand, and finger estimation based on fingerprints using a deep convolution neural network. The publicly available SOCOFIG dataset which embraces 55222 no of fingerprints, is used for training and evaluation of the proposed procedure. On the aforementioned dataset, the suggested mode of operation achieves 99.38% gender, 99.46% hand, and 97.36% finger prediction validation accuracy. The results are competitive and commendable when compared to the preceding techniques.