Comparison of Optimizer on Convolutional Neural Network and Color Representation on Data for Face Presentation Attack Detection
Open Access
- 22 November 2022
- conference paper
- Published by Association for Computing Machinery (ACM)
- p. 121-124
Abstract
Face recognitions have been used for various activities, especially for online verification and security. Face recognition system is a simple biometric, however it is more vulnerable than other biometrics because human face is easy to be manipulated. Face Anti-Spoofing (FAS) system is one of methods for detecting attacks on face recognition system. In this paper, we propose a method for FAS by analyzing the image texture from OULU-NPU database using Local Binary Pattern (LBP) method with Convolutional Neural Network (CNN) as classifier. Our focus is on comparing optimizer on CNN and color representation on the data. The purpose is to find the best optimizer on CNN and the best color representation for FAS system. The FAS model is trained by half of the data from OULU-NPU database which is set in several color representations. The CNN is also set in several optimizers such as Adam, SGD, Adagrad, and RMSprop. The model that is trained in 50 epochs using HSV images with SGD optimizer achieves the best accuracy of 0.99 and area under curve (AUC) of 0.98 among 32 models. From the experiments, it was found that RMSprop optimizer was not suitable for this research.Keywords
This publication has 4 references indexed in Scilit:
- Face Liveness Detection by rPPG Features and Contextual Patch-Based CNN2019
- Replayed Video Attack Detection Based on Motion Blur AnalysisIEEE Transactions on Information Forensics and Security, 2019
- ImageNet classification with deep convolutional neural networksCommunications of the ACM, 2017
- Performance evaluation of texture measures with classification based on Kullback discrimination of distributions2002