This paper presents a robust approach to image tampering detection, utilizing a Kaggle-sourced dataset comprising 6000 images categorized into original and tampered classes. Three deep learning models, VGG19, EfficientNet-B2, and ELA CNN, were employed, with the latter emerging as the best model, achieving an impressive 90% accuracy. To improve overall detection robustness, the suggested methodology combines signature-based, color-based, and metadata analysis techniques. Labeling, scaling, and category encoding are all part of the data preprocessing procedure. Training, testing, and validation data are split 80:10:10. Clear results are obtained by selecting images for tampering predictions through an intuitive graphical user interface (GUI) web application. The integration of CNN and ELA offers a fresh and efficient method for detecting tampering, which is where the innovation lies. The complete and effective method presented in this research ensures the integrity of visual content and makes a substantial contribution to the field. Due to its adaptability, the suggested technique can be used as a useful tool in forensic investigations, journalism, and legal procedures, among other uses, to solve changing difficulties in the rapidly developing digital ecosystem.