Development of an Enhanced Convolutional Neural Network for Fingerprint-Based Ethnicity Identification

Researchers have been extremely concerned in the classification of ethnicity using fingerprint since every individual has distinguished features, that distinguished them based on their ethnicity. The use of fingerprint for identification has been proven to be highly reliable. This research enhanced CNN for fingerprint ethnicity identification system. Fingerprints were collected from three major ethnic groups in Nigeria (Yoruba, Igbo and Hausa), 400 subjects from each of the ethnic groups were collected using Secugen Hamster Plus Fingerprint Scanner. To improve the dataset, image augmentation techniques were applied to 50% of the subject’s fingerprint images collected from three ethnic groups which increased the dataset to 2,400 images which were used for training while the remaining unaugment 50% were used for testing. The raw images were pre-processed; CNN was enhanced using CSO. The performance of the system was evaluated and compare at 0.75 threshold using, Recall (R), F1 score (F1), Recognition Accuracy (RA). The R, F1 and RA were 91.5%, 92.19% and 94.83 for Yoruba, 91%, 91.69% and 94.5% for Igbo and 90%, 91.18% and 94.17% for Hausa for CNN while the corresponding values for CSO-CNN were 94%, 94.7% and 96.5% for Yoruba, 93.5%, 94.20% and 96.17% for Igbo and 94.5%, 95.21% and 96.83% for Hausa ethnicity. CSO-CNN technique performed better than CNN in all metrics. Fingerprint ethnicity identification based on CSO-CNN can be employed to streamline the search range of crime offender by law enforcement authorities where ethnicity is of high concerned.