CNN Image Forgery Detection: Machine Learning Improves Photojournalism Accuracy and Robustness
Varsha Thakur, Rohit Agarwal · 2024
The purpose of this study is to address the rising problem of picture modification and fraud, especially in the field of photojournalism, where the ability to recognize changed photos is essential to preserving faith in digital technology. By employing CNN-based approaches, the created forgery detection system exhibits its capacity to reliably discern between actual and AI-generated photos, reaching a balanced classification of 49% and 50%, respectively, with low misclassification. The purpose of this research is to investigate the development of digital picture forensics by using both conventional methods and contemporary innovations such as blockchain and deep learning. In order to achieve a higher level of detection accuracy, the suggested system makes use of machine learning models, such as transfer learning using pretrained models like ImageNet. The findings demonstrate that the system is capable of generalization and has a limited amount of overfitting, as seen by the high training accuracy of 98.44% and the validation accuracy of 98.92%. It is accepted that there is a need for future improvements in scalability, robustness to adversarial assaults, and interpretability, despite the fact that the findings have been somewhat encouraging. In order to successfully battle increasingly complex picture forgeries, future breakthroughs in digital image forensics will need cooperation between the government, industry, and academic institutions.