YOLO multi-target robust watermarking against cropping attack

Abstract

Digital watermarking is an important technology for secure communication and copyright protection. It is widely used in audio and video anti-counterfeiting protection, identification and traceability. Irreversible desynchronization attacks such as cropping seriously threaten the security and reliability of digital watermarking. To resist cropping attacks, it is an effective scheme to identify multiple foreground targets in an image and repeat the embedding and extraction of watermarks. Yolov5 can quickly identify the location and category of image targets, and the recognition effect is stable. Based on this, this paper proposes a multi-target robust watermarking algorithm based on Yolov5 target detection. First, the main target in the foreground of the image is selected as the watermark to be embedded area through the YOLO target detection network. Then, the discrete cosine transform (DCT) of the transform domain watermark embedding method is selected and the watermark embedding process is completed in the image target detection frame. After the watermarkembedded image has been subjected to different degrees of cropping attack, the residual image is subjected to watermark extraction operation, and the correlation and bit error rate are calculated with the original watermark, so as to verify the robustness of the watermarking algorithm in this paper against cropping attacks. The algorithm in this paper extracts 128 images from the COCOtestval data set for experimental testing. The results show that the method can not only detect the target in the salient region of the image, but also effectively embed a robust watermark, which is an effective solution to digital watermark cropping attacks.