Image Forgery Localization Using U-Net based Architecture and Error Level Analysis
Nagaveni K. Hebbar, Ashwini Kunte · 2021 3rd International Conference on Advances in Computing, Communication Control and Networking (ICAC3N) · 2021
Image forensics mainly deals with checking the authenticity and integrity of the digital images. In this digital era, with the advancements in image manipulation tools and software, the tampering of images for malicious purposes has increased and the detection of such manipulation is necessary. In this paper, an approach using U-Net is proposed to detect multiple types of forgeries and to localize the tampered region at the pixel level in the images. The forged images are processed with Error Level Analysis (ELA) which highlights the forged region having different compression levels and helps in improving the model performance. The processed images are used to train the U-Net based encoder-decoder architecture. A pre-trained Residual Network 50 (ResNet50 backbone network initialized with pre-trained weights is used in the encoder of the network to speed up the training process and enhance the performance of the model by transfer learning. The encoded features are concatenated with the low-level image features at the decoder to localize the tampered region accurately. The proposed approach can detect and localize different types of forgeries with very high precision and outperforms several existing state-of-the-art methods used for forged region localization.