Does this extract the hidden message?

No. It is a detector, not an extractor. It uses statistics to estimate whether sequential LSB embedding is likely present and reports a likelihood, because extraction needs the embedding scheme, ordering and any password.

Why only PNG and BMP, not JPEG?

LSB steganography hides data in the lowest bit of pixel values, and JPEG's lossy compression rewrites those values, destroying the payload. Only lossless formats like PNG and BMP preserve LSBs, so detection is meaningful there.

How does the chi-squared test work?

Natural images have uneven counts within each pair of values that differ only in the last bit (e.g. 100 vs 101). Sequential LSB embedding flattens those pairs toward equality. The chi-squared statistic measures that flattening across all value pairs; a large value indicates likely embedding.

Can it give false positives?

Yes. Heavily processed, dithered or synthetic images can naturally have balanced LSBs and may score high, while data hidden in only part of an image may score low overall. Treat the result as a signal to investigate, not proof.

Is my image uploaded?

No. The image is decoded into a Canvas in your browser and all pixel math runs in JavaScript locally. Nothing is sent to a server, which matters for sensitive forensic material.

Image Steganography Detector

Estimate whether an image hides data

LSB (least-significant-bit) steganography hides a message by overwriting the lowest bit of each colour value — a change invisible to the eye. This tool loads a lossless image (PNG or BMP), reads every pixel via the Canvas API, and applies statistical steganalysis to estimate whether such data is present. It reports a likelihood; it does not extract the payload, because that requires knowing the exact embedding scheme.

How it works

The image is drawn to an off-screen <canvas> and its raw RGBA bytes are read with getImageData. Two complementary tests run over the R, G and B channels:

Chi-squared (Pairs of Values) test. For every pair of adjacent values that differ only in the last bit — (0,1), (2,3), … (254,255) — the test compares the observed counts against the average of the pair. In a clean image these pairs are usually unequal; sequential LSB embedding pushes each pair toward equality. A large chi-squared statistic (and the resulting embedding probability) signals likely hidden data.
LSB-balance check. It counts how many least-significant bits are 1 versus 0. Natural images are rarely exactly 50/50; a ratio extremely close to 0.5 across millions of pixels is a hallmark of an embedded bitstream.

The two signals are combined into a plain-language verdict.

Tips and notes

Use the original file. Re-saving as JPEG, screenshotting, or even some PNG re-encoders will wipe the LSBs and make detection impossible.
A “clean” verdict is not a guarantee — payloads hidden in a small region, or with a non-sequential / encrypted scheme, can evade a whole-image statistic.
A high score warrants deeper forensic tooling (bit-plane visualisation, dedicated extractors) to confirm and recover any payload.