| Abstract |
Preventing cyber assaults is crucial in the setting of a smart city, which is more vulnerable to such threats. Spoofing is a kind of cyber assault that is becoming more widespread across various platforms, including but not limited to email, geo location services, and social media. When someone commits a series of crimes under the guise of another person-for example, fraud, cyberbullying, sextortion, etc.-they are engaging in identity spoofing. A facial recognition system is presented here, with potential use in stopping spoofing. The HOG descriptor, developed by Google, is the basis of this approach. The introduction of entropy would allow us to quantify the significance of each face area in the descriptor, since various face regions do not contain the same information for the recognition process. Therefore, entropy is included to strengthen the algorithm's reliability. When it comes to facial recognition, we've tested our method on three popular databases (ORL, FERET, and LFW), and the results indicate that include entropy information greatly increases the identification rate-by as much as 40% in some of the databases we evaluated. On the CASIA FASD and MIFS databases, spoofing experiments have been conducted, with again improved results compared to comparable texture descriptors techniques. Half Error Rate (HER) is considered as major performance metric and the proposed work minimizes the error up to 1.14. © 2023 IEEE. |
