MSIHNet: A Multi-size Image Hiding Method That Can Hide A Large-Size Image into A Small-Size Image

Image hiding is a task of hiding one image into another without being detected. Current image hiding methods can only achieve the same size hiding, but in real life we may encounter the task of hiding a large-size image into a small-size image. We propose a multi-size image hiding method to solve the problem of image size mismatch. Firstly, we convert the secret image into four small-size sub-bands LL, LH, HL and HH by using discrete wavelet transform. Secondly, according to the different importance of different sub-bands in visual effects and the different hiding effects of different hiding sequences, we hide HH and HL sub-bands into LH sub-band, and then hide LH and LL sub-bands into the cover image. All the hidden networks use the structure of reversible networks. Finally, we recover the secret image by inverse discrete wavelet transform. Experimental results show that our method outperforms other methods on PSNR and SSIM values of the recovered secret image and the stego image in the face of the task that hide a large-size image into a small-size image. Our code can found at: https://github.com/ziencheng3/MSIHNet.