Image Steganography: LSB Technique

A digital colour image is made up of pixels. Each pixel stores three colour channel values: Red, Green, and Blue, each ranging from 0 to 255 and stored in one byte. A 1000x1000 pixel image therefore contains three million individual bytes of colour data. This large volume of data creates room for hiding small amounts of additional information without visible change.

In an 8-bit value, the least significant bit contributes only 1 to the value when set, compared to 128 for the most significant bit. Flipping the LSB of a colour channel changes a pixel value by at most 1 out of 255, a change completely imperceptible to the human eye. This makes the LSB the ideal location for embedding hidden bits without degrading visible image quality.

To hide one byte (8 bits) of secret data, we overwrite the LSB of eight consecutive pixels in a single colour channel. Reading those LSBs back in order reconstructs the original hidden byte. This is the simplest LSB steganography algorithm. The modification to the image is so subtle that pixel pairs like (200, 201) look identical visually while the LSB difference carries meaningful data.

A 24-bit colour BMP image with 1 million pixels has 3 million bytes of pixel data and 3 million individual colour channel values. Using one LSB per channel, we can hide 3 million bits or 375 kilobytes of data. With a carrier file of 10 megapixels, roughly 3.75 megabytes can be hidden. This is sufficient for thousands of text documents or small executables.

The human visual system is much more sensitive to colour hue and saturation than to tiny brightness differences of 1 unit. An image with every LSB replaced by secret data is visually indistinguishable from the original when viewed on a normal display. The perceptual quality metric PSNR (Peak Signal-to-Noise Ratio) for LSB steganography is typically above 50 dB, well above the threshold of human perception.

The simplest approach embeds bits in pixels sequentially from the top-left corner. This is easy to implement and detect: steganalysts know exactly where to look. A more secure approach uses a pseudorandom number generator seeded with a shared key to select which pixels and which colour channels receive hidden bits. An analyst without the key cannot determine where the payload is embedded.

By seeding a PRNG with a secret key, sender and receiver can independently generate the same pseudo-random sequence of pixel coordinates. The sender embeds bits only at those coordinates; the receiver extracts bits from the same coordinates in the same order. An attacker who does not know the key faces a combinatorial search problem even if they know steganography was used.

OpenStego and Steghide are popular open-source tools that implement LSB steganography with optional password-based encryption. Steghide supports JPEG, BMP, WAV, and AU formats and applies a passphrase to both select embedding locations and encrypt the payload. Using a passphrase ensures that even if the presence of hidden data is detected, recovering the content requires the passphrase.

BMP files store raw uncompressed pixel data, making LSB embedding straightforward and lossless. JPEG uses lossy compression that discards high-frequency detail and slightly alters pixel values. LSB embedded in a JPEG before saving will be destroyed by recompression. Steganography in JPEG must work in the frequency domain (DCT coefficients) rather than pixel values, which is more complex but widely used.

Embedding more data increases payload capacity but also increases statistical distortion. Using two bits per channel doubles capacity but makes the statistical signature stronger. Tools like StegExpose can detect common LSB tools by analysing pixel pair distributions. The safe payload is generally kept below 10 to 15 percent of carrier capacity to avoid triggering detection algorithms.

Pixel Pair Analysis exploits a predictable artefact of LSB embedding. Natural images have a characteristic distribution of pixel value pairs; LSB embedding slightly disturbs this by changing odd values to even or vice versa. By measuring the ratio of close-value pixel pairs, steganalysts can estimate both whether embedding occurred and approximately how much data was hidden.

A 24-bit colour BMP image with 1,000,000 pixels uses one LSB per colour channel. How much data can be hidden?

LSB steganography embeds secret bits in the least significant bits of pixel colour channels, causing imperceptible visual change. One byte of hidden data requires eight pixels. A 1-megapixel image can hide about 375KB using one LSB per channel. PRNG-based pixel selection with a shared key improves security. JPEG requires DCT-domain embedding. Statistical analysis via pixel pair distributions can detect LSB embedding.