Proposal of Associative Watermarking Method
Ryoto Kanegae, Masaki Kawamura · 2022 Asia-Pacific Signal and Information Processing Association Annual Summit and Conference (APSIPA ASC) · 2022
We propose a novel type of zero-watermarking method that incorporates associative memory models. Because the zero-watermarking method does not embed a watermark directly in an image, it avoids degrading the original image. However, as a watermark is associated with each image, it is difficult to manage the associations when dealing with a large number of images. Moreover, the conventional zero-watermarking method cannot correct the watermark when an image is degraded, and the watermark's length must be equal to that of the feature vector extracted from an image. Hence, we propose a novel management scheme that uses both a hetero-associative memory model and an auto-associative memory model. The proposed method can manage a large number of mappings between images and watermarks via the hetero-associative memory model, while also eliminating the watermark length restriction. Furthermore, even if an image is heavily degraded, the auto-associative memory model can correct watermark errors. The proposed method was evaluated in the case of JPEG compression, and we found that it can sufficiently reduce errors in watermarking.