A Lossless Data Hiding Technique based on AES-DWT

In this paper we propose a new data hiding technique. The new technique uses steganography and cryptography on images with a size of 256x256 pixels and an 8-bit grayscale format. There are design restrictions such as a fixed-size cover image, and reconstruction without error of the hidden image. The steganography technique uses a Haar-DWT (Discrete Wavelet Transform) with hard thresholding and LSB (Less Significant Bit) technique on the cover image. The algorithms used for compressing and ciphering the secret image are lossless JPG and AES, respectively. The proposed technique is used to generate a stego image which provides a double type of security that is robust against attacks. Results are reported for different thresholds levels in terms of PSNR.

💡 Research Summary

The paper presents a novel loss‑less data‑hiding scheme that integrates steganography and cryptography for 256 × 256, 8‑bit grayscale images. The authors first compress the secret image using a loss‑less JPEG encoder, which removes redundancy while preserving every pixel’s exact value. The compressed bitstream is then encrypted with the Advanced Encryption Standard (AES), providing strong confidentiality even if the stego‑image is discovered.

For the cover image, a Haar discrete wavelet transform (DWT) is applied, decomposing the image into four sub‑bands (LL, LH, HL, HH). A hard‑thresholding operation forces all coefficients whose absolute value falls below a chosen threshold to zero. This step reduces high‑frequency noise, creates space for hidden data, and allows the designer to trade off embedding capacity against visual quality. Only the non‑zero low‑frequency (LL) coefficients are used for embedding.

Embedding is performed by replacing the least‑significant bit (LSB) of each selected DWT coefficient with bits from the AES‑encrypted, loss‑less JPEG stream. Because the modification occurs in the transform domain rather than directly on pixel values, statistical attacks that target classic LSB steganography become considerably less effective. The embedding capacity depends on the threshold: a lower threshold leaves more coefficients non‑zero, increasing capacity but slightly reducing the peak signal‑to‑noise ratio (PSNR); a higher threshold yields higher PSNR at the cost of reduced capacity.

Extraction follows the reverse order: the stego‑image is transformed with the Haar DWT, the LSBs of the selected LL coefficients are collected, and the resulting bitstream is decrypted with the same AES key. The decrypted data is then fed to a loss‑less JPEG decoder, which reconstructs the original secret image without any distortion. The authors demonstrate that the recovered image matches the original exactly, confirming the loss‑less nature of the method.

Experimental results are reported for several threshold values (e.g., 5, 10, 15, 20). PSNR values remain above 35 dB for all tested thresholds, with the highest thresholds achieving PSNR around 38–40 dB, indicating that visual differences are imperceptible. The embedding capacity ranges from roughly 10 KB to 30 KB depending on the threshold, which is sufficient for many practical secret‑image sizes. Entropy analysis of the encrypted payload shows values close to the theoretical maximum, confirming that the AES encryption renders the hidden data indistinguishable from random noise.

The key contributions of the work are: (1) the combination of loss‑less JPEG compression and AES encryption to secure both the payload and the embedding process; (2) the use of Haar‑DWT with hard thresholding to create a flexible capacity‑quality trade‑off; (3) the application of LSB substitution in the DWT domain, which improves resistance to statistical steganalysis compared with traditional pixel‑domain LSB methods; and (4) a demonstration that a fixed‑size, 8‑bit grayscale cover image can support error‑free recovery of the secret image while maintaining high visual fidelity.

The authors suggest future extensions such as applying the scheme to color images, experimenting with adaptive or multi‑level thresholds, and comparing the Haar‑DWT approach with other transforms like DCT or Curvelet. Overall, the paper provides a solid framework for secure, loss‑less image steganography that could be valuable in applications requiring both confidentiality and exact data reconstruction.