Advancing Steganography Through Fuzzy Gradient Selection and Comparative Evaluation

Problem: Contemporary steganography techniques like conventional Least Significant Bit (LSB) and Discrete Cosine Transform (DCT) often exhibit a fundamental trade-off between payload capacity, imperceptibility, and robustness against signal processing attacks. Objective: This work proposes an adaptive steganographic framework that employs intelligent region selection to optimize these trade-offs, minimizing visual artifacts while maintaining data integrity against common distortions. Approaches: We introduce a Fuzzy Gradient Selection (FGS) model based on a fuzzy inference system to select the best regions for embedding within a cover image. The fuzzy membership parameters are then automatically refined to produce optimized embeddings using a Genetic Algorithm (GA). Making the embedding process tailored to each image material. Results: Experimental comparisons indicate that the GA-optimized FGS scheme achieves excellent imperceptibility, registering a Peak Signal-to-Noise Ratio (PSNR) of 65.92 dB, Structural Similarity Index (SSIM) of 0.9999, and only 1.66% of pixels modified at 0.10 bits-per-pixel payload. While payload recovery is ideal under non-attack conditions, the FGS scheme shows considerable resistance to malicious JPEG compression and Gaussian attacks with Bit Error Rates between 0.45 and 0.62 under attack. Conclusion: The results prove that the FGS framework is highly effective when the imperceptibility and statistical undetectability are primary issues. Future work may merge error-correcting codes with spatial-frequency domain methods to enhance robustness while preserving the visual quality achieved by our method.