Résumé
In this manuscript we address data-hiding in images and videos. Specifically we address robust watermarking for images, robust watermarking jointly with compression, and finally non robust data-hiding.The first part of the manuscript deals with high-rate robust watermarking. After having briefly recalled the concept of informed watermarking, we study the two major watermarking families : trellis-based watermarking and quantized-based watermarking. We propose, firstly to reduce the computational complexity of the trellis-based watermarking, with a rotation based embedding, and secondly to introduce a trellis-based quantization in a watermarking system based on quantization.The second part of the manuscript addresses the problem of watermarking jointly with a JPEG2000 compression step or an H.264 compression step. The quantization step and the watermarking step are achieved simultaneously, so that these two steps do not fight against each other. Watermarking in JPEG2000 is achieved by using the trellis quantization from the part 2 of the standard. Watermarking in H.264 is performed on the fly, after the quantization stage, choosing the best prediction through the process of rate-distortion optimization. We also propose to integrate a Tardos code to build an application for traitors tracing.The last part of the manuscript describes the different mechanisms of color hiding in a grayscale image. We propose two approaches based on hiding a color palette in its index image. The first approach relies on the optimization of an energetic function to get a decomposition of the color image allowing an easy embedding. The second approach consists in quickly obtaining a color palette of larger size and then in embedding it in a reversible way.