AI Summary & Key Takeaways
AI InsightExperimental results show that the proposed approach maintains high visual quality (PSNR ≈ 46–59 dB) while providing more stable bit recovery compared to single-coefficient strategies.
📌 3 Key Takeaways
Abstract
In digital image steganography, the primary aim is to embed information imperceptibly while ensuring reliable recovery under degradations encountered during storage, processing, or transmission. Conventional DCT-based approaches are inherently limited, as each bit is typically embedded into a single transform coefficient, making them highly sensitive to distortions such as Gaussian noise, JPEG compression, and resampling. To address this limitation, this study proposes an adaptive multi-coefficient QIM-based embedding framework. In the proposed method, each bit is distributed across multiple mid-frequency DCT coefficients within the same 8×8 block, thereby reducing sensitivity to coefficient-level distortions and improving decision stability. The embedding strength is adaptively controlled based on block variance, enabling a balance between robustness and perceptual quality. In addition, the embedded bitstream is processed using XOR-based pseudo-random encryption and Hamming error correction to mitigate random bit errors. This method is evaluated under Gaussian noise, JPEG compression, and resampling to reflect practical operating conditions. Experimental results show that the proposed approach maintains high visual quality (PSNR ≈ 46–59 dB) while providing more stable bit recovery compared to single-coefficient strategies. The results indicate that it achieves low error rates under JPEG compression and consistent performance under Gaussian noise, whereas resampling remains more challenging due to geometric distortions. The results show that combining adaptive embedding, multi-coefficient redundancy, and error correction improves performance consistency across distortions, although robustness varies with the degradation type.