An Analysis of Image Splicing Detection using Convolutional Neural Network (CNN)

Image splicing is the act of manipulating digital images by copying and pasting pixels from one image onto another, and it can be difficult to detect due to the existence of many editing software. Recently, deep learning approaches have been used to effectively detect image splicing forgery because they learned high-level features from images that are not visible to naked eye. This research aims to evaluate the use of Convolutional Neural Network (CNN) as an image splicing detection method against CASIA V1, Columbia Gray and Columbia Color datasets. The analysis shows that CNN is able to detect 83% accuracy in Columbia Gray, 75% accuracy in CASIA V1, and 69% accuracy in Columbia Color. These results have led to the highest performance, specifically for fake images, which are 96%, 85% and 88%, for each dataset, respectively. This means that the CNN able to detect image splicing correctly, while having difficulties identifying an authentic or real image.