Image-into-Image Steganography Using Deep Convolutional Network

摘要

Raising payload capacity in image steganography without losing too much safety is a challenging task. This paper combines recent deep convolutional neural network methods with image-into-image steganography. We show that with the proposed method, the capacity can go up to 23.57 bpp (bits per pixel) by changing only 0.76% of the cover image. We applied several traditional steganography analysis algorithms and found out that the proposed method is quite robust.