Pendekatan Berbasis GLCM untuk Optimasi Citra Sampul pada Steganografi LSB dengan Keamanan RSA-2048

Abstract

Least Significant Bit (LSB) steganography usually treats every image as an equally good carrier, ignoring that cover texture governs how visible and how detectable the embedded payload becomes. This work proposes a Gray-Level Co-occurrence Matrix (GLCM) approach to rank traditional Indonesian fabric motifs as covers, with the secret first encrypted under RSA-2048 with OAEP padding. Four Haralick features were extracted from 165 grayscale images across 15 motifs, then each image carried multi-bit LSB payloads over a full 3x3 factorial of bit modes and load percentages, yielding 1,485 stego images. Pearson correlation links each feature to quality (MSE, PSNR, SSIM) and to detectability under Chi-square Attack and RS Analysis. Homogeneity is the single strongest predictor, tracking SSIM (r = -0.763 to -0.846) and the RS Gap (r = 0.668 to 0.878), whereas PSNR and MSE capture no texture effect. RS Analysis separates stego from cover far more sharply than Chi-square (AUC 0.986 to 1.000 versus 0.788 to 0.939). A min-max composite score ranks Bali Barong first (0.9394); the worst-case RS Gap between best and worst motif differs by 84.5 percent, confirming that cover choice materially affects security.