SISID: Secret Information Steganography Image Dataset

SISID (Secret Information Steganography Image Dataset) is a comprehensive, reproducible, and synthetic benchmark dataset developed for evaluating image steganography algorithms. The dataset is systematically generated from publicly available benchmark image collections, including BOSSBase, USC-SIPI, Kodak, BOWS2, and DIV2K, ensuring ethical compliance, privacy preservation, and full reproducibility. SISID contains more than 800,000 stego images generated using five embedding algorithms: LSB Replacement, LSB Matching, Adaptive LSB, Index-Based Embedding, and Secret Information-Driven Embedding (SIDE). Each image is evaluated using multiple quality and security metrics including PSNR, SSIM, MSE, Entropy, NPCR, UACI, embedding time, extraction time, and cryptographic payload hashes. The dataset supports multiple payload sizes (128–8192 bits), payload densities (0.05–1.00 bits per pixel), four message formats (ASCII, UTF-8, Binary, and Hexadecimal), grayscale and RGB images, and predefined training, validation, and testing partitions (70:15:15). SISID is intended to facilitate reproducible benchmarking, comparative evaluation, machine learning, deep learning-based steganalysis, and future research in secure image steganography. The complete generation pipeline, metadata, and source code are openly available through GitHub to encourage transparency and community contributions.