Enhancing Image Authenticity Detection: Swin Transformers and Color Frame Analysis for CGI vs. Real Images

This study addresses the increasing difficulty of distinguishing between computer-generated imagery (CGI) and authentic digital images (ADI) due to the rapid advancements in CGI quality. As CGI distinguishment from the real image is becoming increasingly challenging, there is a pressing need for robust classification techniques, particularly in contexts where image authenticity is critical. We have proposed a novel classification framework based on Swin Transformers, incorporating sophisticated preprocessing methods that leverage the analysis of RGB and CbCrY color spaces to improve feature extraction. Unlike traditional machine learning models that rely on handcrafted features, our model directly operates on raw pixel data, which enhances its adaptability and performance. The framework is evaluated using the CiFake-10 dataset, achieving remarkable results: 98.45% accuracy, 97.15% precision, 97.60% recall, and a 97.37% F1-score. Furthermore, we have incorporated complex data augmentation techniques that demonstrate strong resilience against manipulations such as noise addition, blurring, and JPEG compression, making it more generalized. This study shows the effectiveness of Swin Transformers for precise and efficient image authenticity classification in the face of increasingly sophisticated synthetic content.