AI-Synthesized Image Detection: Source Camera Fingerprinting to Discern the Authenticity of Digital Images

This study presents a MATLAB-based framework for detecting AI-synthesized images, specifically focusing on distinguishing real facial images from those generated by artificial intelligence. The system utilizes a deep learning approach, leveraging the ResNet-101 convolutional neural network (CNN) to extract robust features from input images. The image data, obtained from a Kaggle dataset containing real and AI-generated facial images, undergoes preprocessing including resizing and cropping to standardize input dimensions. Features are extracted from the ‘pool5’ layer of the pre-trained ResNet-101 model and used to train a Support Vector Machine (SVM) classifier for binary classification. The dataset is split into training and validation sets to evaluate the model’s generalization capabilities. After training, the system predicts the class of a new input image, identifying it as either a real image or AI-generated. Classification performance is assessed using confusion matrix-based metrics, including accuracy, precision, recall, and F1-score. The approach demonstrated high classification accuracy, confirming the effectiveness of deep CNN features combined with SVM for synthetic image detection. This implementation provides a reliable and scalable solution for digital image authenticity verification, which is increasingly crucial in an era of rapidly advancing generative AI technologies and widespread image manipulation.

Keywords: AI-synthesized images, ResNet-101, support vector machine (SVM), image classification, deep learning.