This PhD project aims at advancing the field of image sensor security through a comprehensive exploration of recent deep learning techniques applied to both robust and fragile invisible watermarking. In the specific context of embedded image rendering pipelines, this study aims to address the dual challenges of ensuring resistance against intentional attacks to break the mark (robust watermarking) while maintaining a high sensitivity to alterations (fragile watermarking). The goal of this multifaceted design approach is not only to enhance the security of imager data but additionally opens avenues for applications in authentication, tamper and forgery detection, combined with data integrity checking. The research will delve into fields of research from image sensor rendering pipeline design using attention-augmented deep learning models to the intricacies of embedding multiple watermarks simultaneously, addressing the requirement for both robust and fragile characteristics. This research is therefore an exciting opportunity for PhD candidates showing interest in the intersection of deep learning, image processing, and security. It provides not only a rich academic landscape for impactful scientific contributions but also holds potential for concrete results for upcoming technological transfers. In practice, the work will consists in finding novel algorithmic solutions to improve watermarking performance, designed to deal with most advanced Deep Learning based attacks, while maintaining a high image quality rendering.