Abstract: Highlights•The exploration of document image forensics in desensitization scenes is pioneered.•An end-to-end contrastive learning-based network, DCLNet, is designed for the task.•A novel dataset, DocTD, is constructed to facilitate network training and evaluation.•Experimental results show DCLNet outperforms existing methods in various cases.
External IDs:dblp:journals/sigpro/Li0ZL026
Loading