Predictive Quality Management in Nepalese A Dual-Level Cuckoo Search Optimization Framework for DNA Sequence Steganography in RGBA Images

Abstract

The growing use of genomic information has increased the need for effective and covert methods to protect sensitive DNA sequence data. Image steganography offers a practical way to conceal such information within digital images while keeping the presence of the hidden data visually imperceptible. In this paper, we propose an adaptive LSB-based steganographic framework for embedding raw DNA sequences into RGBA images. The DNA sequence is first converted into tri-mer indices, providing a compact 6-bit representation. A repetition-aware encoding scheme is then used to handle consecutive identical tri-mers, which reduces the amount of data that needs to be embedded. Depending on the repetition characteristics of the encoded DNA data, the embedding process switches between 3–3 and 2–2–2 bit allocation patterns.To further improve embedding efficiency and stego-image quality, a two-level Cuckoo Search (CS) optimization strategy is introduced. At the first level, CS permutes the tri-mer indices to increase consecutive repetitions, thereby reducing both the effective payload and the number of pixels required for embedding. At the second level, rotation-based optimization improves the match between the secret bitstream and the image LSBs. This reduces the number of bit changes and, consequently, limits embedding distortion. The proposed framework was evaluated using three 512 × 512 benchmark cover images—Lena, Baboon, and Penguin—under three embedding scenarios: direct embedding, repetition-aware embedding without optimization, and the proposed CS-optimized approach. Direct embedding achieved an average PSNR of 44.78 dB with an MSE of 2.16. Repetition-aware embedding increased the average PSNR to 44.90 dB while also reducing the number of pixels used. With CS optimization, the PSNR ranged from 49.07 to 49.12 dB, the MSE decreased to 0.79, and the SSIM reached 0.999932. These results show that the proposed method preserves high visual quality and strong structural similarity between the cover and stego images. In addition, the optimized framework required approximately 95.6% of the pixels used by direct embedding, corresponding to an overall reduction in pixel usage of about 4.38%. Overall, the proposed framework provides an effective approach for reducing embedding distortion and pixel consumption while maintaining high image quality for the concealment of genomic sequence data in RGBA images.