Watermarking is a dynamically developing method of copyright protection used for media (sounds, images, films, or 3D objects) that employs signal processing in order to hide additional, invisible information about the owner or author. However, studies until now have not widely considered the problem of attempting to blind removal of hidden data in a manner that allows the watermarked signal to be returned to the original signal. Current considerations of authors of leading articles in the watemarking field focus on the robustness of the method -security system against intentional attacks of removal of the additional information, without taking into account the aspect of simultaneous degradation of the quality and form of the watermarked picture. As has been shown in the article, it is possible to design a perfect filter (same as reversible operations) that allows the removal of additional information from the watermarked picture in a way that makes it possible to return to the form of the host image. This paper describes an ideal filter used for removal of additional, invisible information in forms of an eliminating function and a signal masking function. Their effectiveness has been demonstrated for practical implementations of this type of eliminating and masking filters for watermarking methods in the cepstrum domain.