IMAGE FORGERY DETECTION USING DEEP CONVOLUTIONAL NEURAL NETWORKS

Abstract

The advent of easily available and simple digital image manipulations has raised issues regarding the validity of the image. Conventional techniques that rely on the hand-crafted feature fail in identifying only certain types of alterations. They also fail to work in practical situations. The current research deals with the issue of developing forgery detection systems based on the analysis of artifacts produced during recompression of images using deep learning. The proposed model, after training with real images and fake images, can detect even minute variations due to double compression and manipulation of images. The pre-processing pipeline was designed completely for improving the quality of images and generating robust features. We evaluated the model with reconjugated real and fake images based on various parameters like accuracy, sensitivity, specificity, and miss rate. The proposed framework attained a training accuracy of 97.38% and validation accuracy of 94.42%, which implies good generalization capability and detection of forgeries. In comparison to other image forgery techniques, the performance of our proposed DCNN framework was found to be quite satisfactory.

Keywords- Image Forgery Detection; Deep Convolutional Neural Network; Digital Image Forensics; Double Image Compression; Recompressed Images; Deep Learning; Image Authentication.