Deepfake Classification For Human Faces using Custom CNN

This paper emphasizes the urgent need for effective deepfake classification methods, particularly for human faces, due to the escalating threat of this technology. The research proposes a comprehensive approach using a Convolutional Neural Network (CNN) model and two Transfer Learning models (ResNet-50 and EfficientNet B7) to address this challenge. It investigates the synthesis of realistic-looking facial manipulations and their societal impacts, highlighting the importance of accurate classification in mitigating these effects. The study evaluates and compares the proposed models' accuracy in detecting manipulated facial content, analyzing their strengths and limitations. Overall, the paper provides a timely exploration of deepfake classification, offering practical solutions to enhance digital security and trustworthiness.