Abstract
<title>Abstract</title> <p>Videos transmitted through social-media platforms are frequently recompressed and cropped. For block-based robust video steganography, recompression perturbs payload-bearing transform coefficients, whereas arbitrary cropping destroys block-grid alignment and shifts both region-of-interest (ROI) coordinates and payload-block locations. To address these coupled distortions, this paper proposes a robust video steganography method combining minimum-embedding-cost adaptive quantization index modulation (MEC-AQIM) with dual-layer pseudo-noise (PN) synchronization. Facial ROIs are detected and tracked, and secret data are embedded in the DWT--SVD domain of the U chroma component. Two PN templates are embedded in selected DCT coefficients of the Y luma component. The periodic fine PN template estimates the intra-block phase offset caused by non-block-aligned cropping, while the aperiodic coarse PN template estimates the block-level displacement after phase alignment. At the receiver, the two estimates are combined to recover the cropping offset, compensate ROI coordinates, and reconstruct the payload block grid before STC and Reed--Solomon decoding. Experiments under H.264, H.265, and VP9 recompression show that the proposed method largely preserves the recompression robustness of MEC-AQIM, achieves synchronization success rates of 93.5%--97.0% for the tested cropping offsets at 720p and 1080p, and reliably recovers complete messages under most combined attacks while maintaining high visual fidelity and a usable payload capacity.</p>