In this paper, a new methodology for the secure embedding of data in a video sequence is presented. To guarantee the imperceptibility of the embedded data, we propose a method to select frame regions that are considered perceptually non relevant. For each frame a salience analysis is performed based on features that are relevant to the Human Vision System. In particular, the local contrast, the color and the areas of motion have been considered. By weighting all these feature at once, an importance map is built to drive the sequent embedding procedure. Subjective experiment results show that the artifacts caused by this localized embedding procedure are considered less annoying than if the embedding is performed on the whole frame. Nevertheless robustness is achieved.