Deeply‐Recursive Attention Network for video steganography

Video steganography plays an important role in secret communication that conceals a video cover by perturbing the value of pixels frames. Imperceptibility is first and foremost requirement any steganographic approach. Inspired fact human eyes perceive pixel perturbation differently different areas, novel effective efficient Deeply-Recursive Attention Network (DRANet) for to find suitable areas information hiding via modelling spatio-temporal attention proposed. The DRANet mainly contains two components, Non-Local Self-Attention (NLSA) block Co-Attention (NLCA) block. Specifically, NLSA can select frame which are computing correlations among inter- intra-cover NLCA aims effectively produce enhanced representations frames enhance robustness model alleviate influence video. Furthermore, reduces parameters performing similar operations on within input recursively. Experimental results show proposed achieves better performance with fewer than state-of-the-art competitors.