摘要

High-accuracy image steganography seeks to conceal secret binary messages within a single cover image and recover them with minimal error. However, the pursuit of accuracy often introduces distortions, such as noise and artifacts, increasing the risk of detection. To cope with this issue, we propose an innovative Frequency-guided Iterative network for secure image Steganography, namely FIS, to minimize detectability while preserving visual quality in this paper. FIS comprises four key components. First, a flexible iterative encoder, implemented as a recurrent neural network, refines the cover image iteratively to identify optimal pixel locations for data embedding. Second, a frequency perturbation module strategically directs the hiding of secret information into the high-frequency domain, effectively reducing detectability. Third, a decoder incorporates a secure mechanism to obfuscate the output, providing robust protection against unauthorized access. Finally, we introduce a lightweight and effective attention module that enhances feature selection throughout the network. Experimental results across multiple datasets demonstrate the superiority of FIS over other state-of-the-art methods, showcasing its improved recovery accuracy, hiding invisibility, and overall security.