摘要

In this paper, we present a method for detecting and restoring tampered information in natural and text images. To take the detection ability, invisibility, and recovery quality into account for both natural and text images, this work generates the authentication watermark for each 4 x 4 block by a hash function and a variable capacity recovery watermark for each 2 x 2 block by allocating more bits to the textural blocks and fewer bits to the smooth ones. The authentication watermark and the recovery one are embedded in the original image by adopting different strategies based on a secret key. The multi-stage neighbor detection strategy is designed to locate the tampered image blocks accurately. The proposed scheme outperforms in invisibility, with detecting tampered locations and recovery of the tampered regions. The simulation results show that the proposed scheme achieves better recovery quality and invisibility for natural and text images due to a variable-capacity recovery watermark and superior localization. Further, the proposed method computationally is less expensive compared with the existing works.