<![CDATA[Steganography serves as a critical information security technique for concealing data within digital media. While the spatial-domain Least Significant Bit (LSB) method is widely adopted due to its embedding effectiveness and straightforward implementation, this study addresses a crucial gap: the lack of implementation-level comparisons of deployable Python LSB libraries utilizing dual-metric evaluation and standardized robustness testing. We present a systematic comparative performance analysis of three distinct Python-based implementations: classical LSB sequential, LSB randomized, and Discrete Cosine Transform (DCT)-based LSB embedding. Image quality and fidelity were rigorously quantified through Peak Signal-to-Noise Ratio (PSNR) and the Structural Similarity Index Measure (SSIM). Under ideal conditions, baseline evaluations demonstrated high imperceptibility across all methods, yielding PSNR values ranging from 43.1 to 48.2 dB and SSIM scores between 0.85 and 0.95. However, standardized robustness testing by encompassing Gaussian noise, spatial cropping, and rotational manipulations exposed significant vulnerabilities. Post-manipulation image quality assessments revealed severe structural degradations, with PSNR values dropping to a range of 6.81 dB to 22.79 dB and SSIM scores falling between 0.6454 and 0.8781, depending on the attack type. Consequently, classical LSB methods exhibited Bit Error Rates (BER) of 44-54% for color images and 45-50% for grayscale images. Notably, the DCT-based method demonstrated superior resilience against geometric transformations, significantly reducing the BER to 25.37% under rotational attacks for grayscale images, compared to 50% for classical LSB. These findings provide vital empirical guidance for selecting appropriate Python implementations based on specific application requirements, effectively balancing embedding capacity, imperceptibility, and robustness against attacks.]]>
