What is a file hash (checksum)? Proving two files are identical

Which hash functions are common?

For spotting duplicate photos on a phone, speed matters more than resistance to attackers, so tools often use a fast hash, sometimes only over part of the file, then confirm a match with a full comparison.

A hash is a one-way calculation. You cannot rebuild a photo from its SHA-256 value, and the value is the same length, 64 hexadecimal characters, whether the file is a 50 KB sticker or a 50 MB RAW.

Download sites publish checksums for the same reason: after a large download, computing the hash yourself and comparing it shows whether the file arrived intact.

How does a duplicate finder use hashes?

1. List every image file and its size. Files of different sizes cannot be identical, which rules out most pairs instantly.
2. For files that share a size, compute a hash of the contents.
3. Group files with matching hashes. Each group is one image stored several times.
4. Keep one copy per group, usually the original in the camera folder, and offer the rest for deletion.

Because the work is simple arithmetic, it runs entirely on the phone and takes seconds even for tens of thousands of files. GalleryJury, coming soon to Android, finds exact copies this way on the device before any AI step begins. The full approach is covered in the guide to deleting duplicate photos on Android.

Hashing also makes re-runs cheap. A tool can store each file's hash and skip unchanged files next time, so only new photos need work.

What can a file hash not detect?

Anything that changes the bytes, even when the picture looks the same. A WhatsApp copy is re-compressed, an edited file has new pixels, and a copy with stripped metadata has fewer bytes. All three get unrelated hashes. Finding those needs a perceptual hash or a visual model, which is the gap between a simple exact duplicate check and a similar-photo tool.

Two different files with the same hash, a collision, is theoretically possible. With SHA-256 the odds are so small that no accidental collision has ever been found, so a match is treated as proof in practice.

Metadata edits catch people out. Rotating a photo in some galleries only changes the orientation tag, yet that alone produces a new hash, so the rotated file and the untouched original stop matching.