A Study of Various Steganographic Techniques Used for Information Hiding

1. INTRODUCTION

• Digital images are the most popular carrier files in the current situation, among applications ranging from military use to medical imaging.The cover medium is chosen with the secret message's type and size in mind.Source: §1 · p. 1
• The paper examines embedding techniques and their advantages and differences, with the goal of concealing secret data in a cover.The intended result is that only the sender and recipients realize secret data is present.Source: §1 · p. 3
• A stego-medium is expected to be imperceptible and robust, and embedding should not reduce stego-image quality.Imperceptibility means observers cannot distinguish the original from the stego-image; robustness concerns withstanding attacks.Source: §1 · pp. 2–3

2.1. Classification of Steganographic Categories

• Steganography is divided into pure, secret-key, and public-key categories according to key use.Pure steganography uses no key; secret-key steganography exchanges a key beforehand, while public-key steganography uses public and private keys.Source: §2.1 · p. 3

2.2 Classification of Steganographic Methods

• The six method categories are substitution, transform domain, spread spectrum, statistical, distortion, and cover generation.The paper notes that exact classification is not always possible.Source: §2.2 · pp. 3–4
• Substitution replaces redundant cover parts, transform-domain methods embed in signal transform space, and spread-spectrum methods borrow communication ideas.Statistical methods alter cover statistics for hypothesis testing; distortion methods encode through signal distortion, while cover generation encodes through how the cover is created.Source: §2.2 · pp. 3–4

3.1 Substitution Methods (Spatial-Domain)

• Spatial-domain proposals include 3D mesh embedding resistant to cropping, rotation, and scaling, pixel-difference and LSB embedding, and interleaved prediction for reversible hiding.The 3D method embeds in newly added mesh positions; the predictive method keeps pixel differences within ±1.Source: §3.1 · pp. 4–5
• The randomized RGB-pointer method is reported to provide better secrecy and performance with enhanced embedding capacities than its two alternatives.It uses one channel’s LSBs to indicate embedding in the other two, with pointers chosen cyclically in the randomized method.Source: §3.1 · pp. 5–6
• The edge-region LSB method reports improved distortion performance and resistance to steganalysis, while its non-adaptive variant has higher PSNR and lower MSE than its adaptive variant.Embedding uses paired pixels and a threshold to select sharper edge regions.Source: §3.1 · pp. 4, 8
• The encrypted, noise-filtered method combines error detection and correction with reported noise-free transmission and a success rate above 83% across different image types.It substitutes secret bits into pixel octets and uses ARQ after extraction.Source: §3.1 · pp. 5, 8
• A movie-clip carrier separates static and dynamic portions for embedding and is reported to provide very high capacity with security.The static portion stores three characters per pixel, while the dynamic portion uses an MSB method and a separate stego-key.Source: §3.1 · pp. 6, 9
• The cross-fold transposition and genetic-algorithm method is reported to exceed S-Tools in text embedding capacity and security.It perturbs the secret text’s binary form with a genetically generated session key before embedding it in the host image.Source: §3.1 · pp. 7, 9

3.2. Transform Domain Methods

• Transform-domain studies use Integer Wavelet Transform to hide secret images in color covers, reporting better stego-image quality in RGB than YCbCr and hiding multiple images and keys in YCbCr.The first study compares RGB and YCbCr embedding; the second encrypts two keys before hiding them in the cover image.Source: §3.2 · p. 10
• Haynes uses computer vision, machine learning, and Haar features so image classification yields a hidden message, with DWT aiding concealment in ordinary unmodified images.The proposed message cannot be decrypted without key compromise.Source: §3.2 · p. 10
• Arivazhagan et al. apply Arnold Transform and DWT to a secret colour image before embedding its concatenated RGB bitstream into a DWT-processed colour cover, aiming for robust extraction with minimal cover changes.Their modified colour-image embedding approach addresses limitations of previous methods.Source: §3.2 · pp. 10–11
• Sarkar et al. use ME-RA coding to convey hidden bits through host-coefficient syndromes, reporting improved steganalysis performance over QIM in YASS but greater software complexity; Gope et al. encrypt JPEG-block data using CRC checking.ME-RA uses pseudo-random hiding blocks and repeat-accumulate error correction; Gope’s method applies DCT and quantization to 8 x 8 blocks.Source: §3.2 · p. 11
• Po-Chyi et al. describe JPEG 2000 wavelet and Lazy Mode embedding, while Noda et al. combine JPEG 2000 with BPCS block replacement and conjugation maps.JPEG 2000 can carry high-volume data, but modifying its compactly compressed packets risks preventing image expansion; Noda’s approach embeds after tier-2 encoding.Source: §3.2 · p. 11

3.3. Statistical Methods

• Filler et al. use Syndrome-Trellis Codes and a chosen distortion function to construct flexible spatial- or transform-domain embedding schemes that hinder statistical detection.The distortion function and embedding operations need not be shared with the recipient.Source: §3.3 · p. 13
• Fridrich et al. propose reversible embedding in VQ-compressed images using side matching and relocation, without a location map; secret bits embed only in one codeword case.The method partitions the codebook into clusters for imperceptibility and uses compensation or skips other codeword cases.Source: §3.3 · p. 13
• Chang et al. combine wet-paper codes, matrix embedding, and random linear codes to support arbitrary selection channels with fewer embedding changes; Ni et al. use block differences for lossless hiding robust to JPEG/JPEG 2000.Ni et al. embed using threshold shifts of the block difference value and apply error correction.Source: §3.3 · pp. 13–14

3.4. Distortion Methods

• Ould MEDENI et al. discuss matrix encoding with an agreed parity-check matrix, while Gaikwad et al. blur a secret image before LSB embedding and recover it using PSF parameters and keys.Gaikwad et al. encrypt the message dimensions in the cover’s first row and send secret key values through a secure channel.Source: §3.4 · p. 15

• The study reviews articles on steganography and categorizes them into different techniques.Source: §4 · p. 15
• Named application areas include internet banking, mobile communication security, and cloud security.Source: §4 · p. 15
• The authors suggest steganographic principles can guide identification of new areas and improvement of existing applications.Source: §4 · p. 15