A Deep Learning Approach to Invisible Watermarking for Copyright Protection

Watermarking has been an age-old process serving a multitude of purposes from copyright protection to maintaining confidentiality. Invisible watermarking is a subcategory of watermarking which involves the use of a digital signature embedded in the image to be protection in such a way that keeps the watermark imperceptible to the naked eye. The proposed technique embeds an invisible and ineffaceable watermark to any image, without any distortion or loss in image quality. This is done using autoencoder neural network which is trained for the process of embedding the watermark onto the image to be protected. This technique provides a lower loss after extraction and makes sure the watermark is invisible with minimal distortion to original image. Easy extraction of watermark is also ensured using the model specifically designed for extraction.