Majority of images is widely spread in the virtual universe of online based social media entertainment. With the accessibility of numerous altering programming tools that permits us to alter images, evidently in various image processing applications there exist unidentified numerous falsification images. Scientifically using the error level analysis we can find out the pressure proportion between the first image and the phony picture, in light of the fact that the first image pressure and phony images are unique. Besides knowing whether the image is authentic or counterfeit we can drill down the metadata properties of the image, however the modification of metadata can be easily performed. Under these circumstances we apply deep learning to perceive controlled images via preparing the model utilizing dataset containing phony and genuine images through adopting error level analysis strategy on each image and fine tuning various parameters for accurate error rate analysis. In our research work random forest algorithm and CNN technique are compared for the effectiveness of image forgery detection task. Finally, the proposed CNN method witnesses the best precision of almost 99% compared to random forest algorithm for 50 epochs upon custom dataset consisting of 200 real and 100 fake images.

CNN, Deep Learning, ELA, Random Forest, Image Forgery