A digital watermark is a message robustly hidden within another digitized signal such as an image, a piece of music, a video, etc. . The principal applications of this technology are copyright protection and document fingerprinting. The robustness of watermarking algorithms against common geometrical deformations has drawn the attention of many researchers in the last ten years. Such distortions can result from usual processing of the media and/or malicious manipulations. As an example, a major concern for digital cinema industry is the illicit copy of movies with video cameras taking place in movie theaters. In this scenario, severe geometric distortions can jeopardize the retrieval of the watermark message from the illicit copies. The limitations and weaknesses of the previously proposed solutions to fight these distortions are presented. We developed a generalized construction method for periodic pseudo-random patterns. Based on these patterns, we designed a spread spectrum watermarking scheme with enhanced security properties. We investigated the detection probability and the interaction between exhaustive search and informed coding strategies. Finally, a scheme for the detection of a periodic structure and for the inversion of affine distortions was presented. We showed that the choice of the periodic repetition size involves a trade-off between robustness and secrecy. Thereafter, we studied the security flaw caused by the lack of secrecy in pilot-registration approaches. We proposed an innovative hiding scheme to remedy this issue. Our solution involves the extraction of robust local references from the content of the cover signal. Using this content normalized interpretation, we showed how one can design robust secret binary mask and modulate pilot signals in watermarking schemes. The efficiency of the approach is demonstrated on pilots derived from periodic structures. We also addressed the assessment of the degradation introduced by a geometrical distortion. We assume that a global rigid transform does not impair the perceptual value of an image and we propose a new criterion based on a local analysis of the geometrical deformations.