Image Watermarking Technique Using Convolutional Autoencoder

Due to the rapid growth in communication networks and data sharing, protecting copyrights has become increasingly important. Digital watermarking is seen as a promising solution to resolve this issue. A new blind digital watermarking based on deep convolutional auto-encoders is presented in this paper. The system facilitates the embedding and extraction of a watermark in an image for its protection. The proposed architecture comprises four modules: an “encoder” that compresses the original image and watermark, an “embedder” that combines the two, a “decoder” that upsamples the combined result to reconstruct the watermarked image, and an “extractor” for watermark retrieval. The experimental results indicate that the watermarked images maintain their visual quality after insertion, and the extracted watermark remains robust against various various attacks such as filtering, rotation, and noise.