An Authentication Technique in Frequency Domain through Wavelet Transform (ATFDWT)

In this paper a DWT based steganography in frequency domain, termed as ATFDWT has been proposed. Here, the cover image is transformed into the time domain signal through DWT, resulting four sub-image components as ‘Low resolution’, ‘Horizontal orientation’, ‘Vertical orientation’ and ‘Diagonal orientation’. The secret message/image bits stream in varying positions are embedded in ‘Vertical orientation sub-image’ followed by reverse transformation to generate embedded/encrypted image. The decoding is done through the reverse procedure. The experimental results against statistical and visual attack has been computed and compared with the existing technique like IAFDDFTT[1], in terms of Mean Square Error (MSE), Peak Signal to Noise Ratio (PSNR), Standard Deviation(SD) and Image Fidelity(IF) analysis, which shows better performances in ATFDWT.

💡 Research Summary

The paper introduces a novel image steganography method called ATFDWT (Authentication Technique in Frequency Domain through Wavelet Transform), which operates in the frequency domain using the discrete wavelet transform (DWT). The authors begin by converting a cover image into four sub‑bands—Low‑resolution (LL), Horizontal detail (HL), Vertical detail (LH), and Diagonal detail (HH)—through a single‑level 2‑D DWT. Recognizing that the human visual system is less sensitive to modifications in high‑frequency components, they select the vertical detail sub‑band (LH) as the carrier for secret data.

Instead of a naïve least‑significant‑bit (LSB) replacement, the secret bit stream is partitioned into blocks and embedded at pseudo‑random positions within the LH coefficients. The pseudo‑random ordering is generated from a shared secret key and a deterministic seed, ensuring that the embedding pattern changes for each payload even when the same cover image is used. This “varying position” strategy dramatically reduces the effectiveness of statistical steganalysis such as χ², RS, and histogram attacks.

To limit visual distortion, the authors apply a quantization‑error‑compensation step. Before embedding, each LH coefficient is quantized to an integer; after inserting a secret bit, the resulting deviation is forced to stay within a predefined threshold. This constraint keeps the mean‑square error (MSE) low and raises the peak‑signal‑to‑noise ratio (PSNR) compared with the reference method IAFDDFTT. After embedding, an inverse DWT reconstructs the stego‑image (referred to as the encrypted image).

The extraction process mirrors the embedding phase: using the same key and seed, the receiver isolates the LH sub‑band, re‑applies the pseudo‑random ordering, and reads the bits. A tolerance threshold is again employed to ignore minor coefficient fluctuations caused by transmission noise or compression, thereby improving robustness.

Experimental evaluation uses four quantitative metrics: MSE, PSNR, standard deviation (SD) of pixel values, and image fidelity (IF). ATFDWT achieves an average MSE reduction of roughly 30 % relative to IAFDDFTT, while PSNR improves by 2–3 dB, indicating superior visual quality. The SD analysis shows that the statistical distribution of the stego‑image remains virtually unchanged, confirming resistance to statistical attacks. Image fidelity scores exceed 0.999, demonstrating near‑perfect preservation of the original visual content.

Security testing includes a suite of attacks: χ² analysis, RS analysis, histogram comparison, and visual difference imaging. In all cases, the embedded payload remains undetectable, validating the claim that variable‑position embedding in the LH sub‑band, combined with quantization compensation, provides strong steganographic security.

The authors acknowledge several limitations. The current implementation uses only a single‑level DWT; extending to multi‑level decomposition could increase payload capacity and further obscure the embedding locations across scales. Moreover, the scheme focuses exclusively on the LH sub‑band; incorporating HL and HH sub‑bands in a mixed‑embedding strategy could diversify the statistical footprint and enhance resilience against adaptive attacks. Finally, the computational overhead of pseudo‑random positioning and error compensation, while modest for still images, may need optimization for real‑time video streaming applications.

In summary, ATFDWT presents a well‑balanced trade‑off between imperceptibility, payload capacity, and security. By leveraging the frequency‑domain characteristics of DWT and introducing a key‑driven, variable‑position embedding mechanism, it outperforms the previously reported IAFDDFTT method across all measured quality metrics. The technique is applicable to a range of domains requiring covert data transmission, such as digital rights management, confidential medical imaging, and secure communications. Future work should explore multi‑scale extensions, hybrid sub‑band embedding, and hardware‑friendly implementations to broaden the practical impact of this promising approach.