Vgg-16 And Vgg-19 Architecture Models In Lie Detection Using Image Processing

Abstract

The polygraph, LVA, and VSA are three techniques used to find lying. However, the polygraph monitors the physiological reactions that have been linked to lying rather than detecting lies. Validity and reliability testing for VSA and LVA has been conducted, but the results are not reliable enough and unsatisfactory. How to find a lie detection model using the CNN approach is the goal of this research. With the addition of an augmentation process to image processing, dropouts, and three optimization algorithms-Adam, RMSprop, and Adadelta on the two CNN architecture models with VGG-16 and VGG-19 we use. For the dataset this study used a video-based public dataset, the video was converted into an image format, and a confusion matrix was used to analyze the data during validation. The study found that the accuracy achieved from the VGG-16 model that is, Adam + VGG-16 has the best models were obtained from 18 combinations combining the VGG-16 and VGG-19 models along with other optimizations.