Image Hiding and Restoration via Deep Moiré Networks.

Moiré patterns arise when two repetitive textures are superimposed, which often degrade the image quality when taking photos. However, controllable moiré patterns can be useful in many practical applications e.g. image hiding. Existing moiré-based approaches heavily rely on manually designed filters, resulting in slow and inflexibility. Moreover, existing methods often struggle to recover image details, making it challenging to achieve satisfactory visual quality. To address these limitations, in this study, a deep moiré hiding network (DMHN) is proposed to control the generation of moiré patterns and applied to the image hiding task. The network encodes the secret image to generate modulation parameters of moiré patterns and hide the images in unreadable gratings. When superimposing the hidden image with the key grating, the hidden image becomes visible again in the form of moiré patterns. To better restore the moiré-based reappeared images, a grating removal network (GRN) is also presented to eliminate the residual gratings and recover high-frequency image details. Experiments demonstrate that the proposed method can effectively hide and restore the images via moiré modulation and generation.