Evaluate the Performance of Deep CNN Algorithm based on Parameters and Various Geometrical Attacks

This paper uses learning-based color illumination techniques for the classification and localization of passive image geometric attack forgery classification and localization. Copy move and splicing forgery are classified using CNN. The classification accuracy obtained during validation for CASIAv1.0 is 97.35, CASIAv2.0 is 97.93, and DVMM is 97.86. The large data set is created to classify geometrical attacks by combining all the data sets with rotation and scale artifacts. The classification accuracy between rotation and scale during validation is 99.29. Machine learning-based color illumination technique is used for localization of forgery. An experiment was conducted on the CoMoFoD data set to detect passive image and geometric attack forgery. There are 48 images in the dataset with various geometric attacks such as scale and rotation. The results for identifying simple CMF attacks show an F1 score of 98.53%, a precision rate of 97.25%, and a recall rate of 100%. In the case of detecting CMF attacks on a larger scale, the F1 score is 79.1%, the precision rate is 85.2%, and the recall rate is 74.8%. For CMF attack rotation, the F1 score is 86.16%, the precision rate is 87.83%, and the recall rate is 76.33%. The proposed method demonstrates improved accuracy in detecting forgeries compared to existing approaches.