Zhang, Y. & Thing, V.L.L. Multimed Tools Appl (2018). https://doi.org/10.1007/s11042-018-5756-4
In multimedia security, it is an important task to localize the tampered image regions. In this work, deep learning is used to solve this problem and the approach can be applied to multi-format images. Concretely, we use Stack Autoencoder to obtain the tampered image block features so that the forgery can be identified in a semi-automatic manner. Contextual information of image block is further integrated to improve the localization accuracy. The approach is tested on a benchmark dataset, with a 92.84% localization accuracy and a 0.9375 Area Under Curve (AUC) score. Compared to the state-of-the-art solutions for multi-format images, our solution has an over 40% AUC improvement and 5.7 times F1 improvement. The results also out-perform several approaches which are designed specifically for JPEG images by 41.12%∼63.08% in AUC and with a 4∼8 times better F1.
This material is based on research work supported by the Singapore National Research Foundation under NCR Award No. NRF2014NCR-NCR001-034.
This is a post-peer-review, pre-copyedit version of an article published in [insert journal title]. The final authenticated version is available online at: https://doi.org/10.1007/s11042-018-5756-4