Image Steganography Embed & Detect

Embedding writes UTF-8 bytes into the LSBs of each pixel's channels in the interoperable order (row-major, R→G→B, MSB-first — the same convention CyberChef's “Extract LSB” and StegOnline read), framed by a 12-byte STG1 header (magic + flags + 32-bit big-endian length) so extraction is self-describing and clean images almost never false-positive. A scatter seed switches to a seeded Fisher–Yates permutation of the sample stream (cyrb128 + mulberry32): sequential tools then read only noise, and the chi-square falling curve flattens. Transparent pixels are skipped by default because browsers' premultiplied-alpha pipelines treat fully-transparent RGB as don't-care and can destroy hidden bits on re-save; the output PNG is bit-exact elsewhere (LSB replacement changes each touched sample by at most 1). Detection computes the chi-square statistic over PoV pairs (2k, 2k+1) per channel — embedded data equalizes each pair's histogram — with the p-value from the regularized incomplete gamma function, applied over sliding prefixes (1%…100%) to localize where a sequential message ends, plus an LSB bit-plane preview and LSB entropy. Educational/CTF scope: this is not encryption, and scattering's 32-bit seed is brute-forceable.