Image Forgery Detection using ELA and CNN
Sanika S Patankar, Ashutosh Joshi, Gaurav Durge, Atharva Jaid, Khomesh Kalambe, Hrishikesh Dhale · 2023
Picture tampering has recently increased due to the increasing popularity of photo-editing software, making it difficult to determine the authenticity of an image with the naked eye. To address this issue, we proposed a model that detects image forgeries using ELA (Error Level Analysis) and CNN (Convolutional Neural Network) techniques. ELA is a forensic tool that examines variations in compression levels, whereas CNN is a deep learning technique that can learn and extract features from photos to categorize them as either modified or legitimate. The model combines both techniques and was tested on the publicly available CASIA 2.0 dataset. According to the results, the proposed model outperforms existing deep learning models in terms of training time and computational efficiency, achieving an overall accuracy of more than 95.19%. The proposed method is an effective and efficient method for detecting a wide range of image editing techniques used for image manipulation.