Deepfake Image Forgery Detection for Suspicious Images

Deepfake technology has emerged as a potent tool for creating realistic yet fabricated images that pose significant challenges to the veracity of visual content. This paper proposes an advanced framework for the detection of deepfake image forgery, aiming to identify suspicious images with a high degree of accuracy. The approach integrates a multi-faceted analysis, incorporating digital forensics, deep learning techniques and image quality metrics. Using the power of deep learning networks—more specifically, XceptionNet and a GAN network that makes up EfficientNet—this paper presents a novel method for deepfake image forgery detection that can effectively discriminate between real and manipulated images for both low and high quality AI-synthesized images. Low-quality images are those which are generated through face swap and face reenactment while high quality images are those which are generated through face synthesis. We're grouping into classes that are real, fake and suspicious, suspicious for those which are false positives for the real and fake classes. Lastly, additional analysis is conducted on the fake and suspicious classes in order to draw appropriate conclusions. Our experimental findings show that the suggested method for deepfake image forgery detection performs better than current approaches. The proposed method achieves 95% of accuracy rate on a dataset consisting of both real and fake images. This implies that the technique might prove to be a useful resource for identifying deepfakes in practical settings.