Locate-then-Reveal: Enhancing the Playload Capacity of Multi-image Hiding

Multi-image hiding aims to enhance image hiding capacity by embedding different secret images within a single cover image, facilitating efficient message transmission. Existing work focuses on encoding different secret images into secret codes of the same size and fusing them in the channel dimension to obtain a composite code that contains all secret information. However, these methods struggle with larger numbers of secret images because the fusion in the channel increases the difficulty of secret separation, which causes significant degradation in recovered image quality. In this study, we introduce a novel Local Deep Hiding (LDH) scheme for multi-image hiding, achieving a breakthrough by hiding over 10 full-size secret images within a single cover image when existing methods can only hide 2 or 3 images within an image. Unlike prior methods, LDH employs a compact encoding strategy using a newly designed hiding network, reducing the secret code size to only \(\frac{1}{16}\) of the original image. This enables easy concealment within local regions of the cover image, with the remaining regions available for additional secret images. To recover hidden images, LDH employs a locating network to identify secret code locations within stego images, followed by a revealing network to reconstruct full-size secret images independently. Experimental results demonstrate LDH’s superiority in payload enhancement, maintaining a Peak Signal-to-Noise Ratio (PSNR) of over 33 dB between the secret images and their recovered versions, even with up to 16 secret images.