Deep Learning Technique Application for The Detection of Synthetic Images

Abstract

This study presents a novel approach for detecting synthetic images using a Convolutional Neural Network (CNN). The proposed approach makes use of a two-step procedure: first, data collection and preprocessing; next, model training and assessment. While pre-trained diffusion and Generative Adversarial Network (GAN) models were used to create synthetic images, real images were used from publicly available datasets such as FFHQ, AFHQ, and LSUN. To differentiate between genuine and artificial photos, a CNN model with a complex architecture was created. It consists of fully connected layers for classification and convolutional layers for feature extraction. Using a 10-fold cross-validation method, the system's average accuracy, precision, recall, and F1- score were 96.7%, 0.96, and 0.97, respectively. The results obtained show how well the model detects synthetic images with high recall and precision, underscoring its potential for practical uses in content authentication, digital forensics, and AI-generated image recognition. The study emphasises how crucial it is to use deep learning methods to tackle the escalating difficulties in synthetic image identification.