deep-recover

Recover deleted files from disk images (USB sticks, SD cards, hard drive images) using two complementary engines:

1. Metadata recovery — parses the actual filesystem structures (FAT directory entries, NTFS MFT, ext2/3/4 inode table) via pytsk3 to find entries still marked "deleted" whose data hasn't been overwritten yet. Gives you the real filename, size, and timestamps.
2. Signature carving — scans raw bytes for file "magic numbers" (carver.py), independent of any filesystem metadata. Works even if the filesystem index is damaged or the drive was reformatted, at the cost of losing filenames and sometimes exact file boundaries.

deep-recover runs both by default and de-duplicates results by content hash.

⚠️ This tool only works on data that hasn't been overwritten. "Delete" in almost every filesystem just removes the pointer to a file — the data itself stays on disk until something else reuses that space. See How this works below. On SSDs with TRIM enabled, that window can close within seconds of deletion — see Limitations.

Legal / ethical use

Only run this against media you own or have explicit written authorization to examine (e.g. your own USB stick, a drive you're contracted to investigate). Recovering data from a device you don't have the right to access is illegal in most jurisdictions, full stop. This tool does not defeat encryption, does not touch remote/network systems, and does not do anything a live human forensic examiner couldn't do by hand with a hex editor — it just automates it.

Installation

git clone https://github.com/Bernard411/deep-recover.git
cd deep-recover
pip install -r requirements.txt --break-system-packages   # Kali/Debian externally-managed envs
pip install -e .

pytsk3 (used for metadata recovery) needs build tools on some systems:

sudo apt install build-essential python3-dev libtsk-dev
pip install pytsk3

If pytsk3 fails to install, deep-recover still works — it just falls back to carving-only mode automatically.

Quick start

Step 1 — never run this against a live device. Image it first:

lsblk                      # find your USB stick, e.g. /dev/sdb
sudo umount /dev/sdb1       # unmount, don't format
sudo dd if=/dev/sdb of=usb_image.dd bs=4M status=progress

Step 2 — run deep-recover against the image, not the device:

deep-recover usb_image.dd -o ./recovered

Output:

deep-recover -- source: usb_image.dd
Output directory: ./recovered

[metadata] Found 3 deleted-but-intact file entries.
[carve] Scanning usb_image.dd for file signatures ...
[carve] Scanned 64 MB, found 2 files:
           pdf     : 1
           png     : 1

Done.

Recovered files land in ./recovered/, named either after their original filename (metadata engine) or <type>_<offset>_<hash8>.<ext> (carving engine).

Usage

deep-recover SOURCE [-o OUTPUT] [--mode {auto,metadata,carve}] [--types TYPE [TYPE ...]] [--offset BYTES] [-v]

Supported carve types: jpg, png, gif, pdf, zip (also matches docx/xlsx/pptx/apk/jar, since they're zip containers), gzip, bmp, mp4, sqlite, wav. Add more in deep_recover/signatures.py.

How this works

Why "delete" doesn't delete

Deleting a file removes its directory/index entry (and, on FAT, flips the first byte of the filename) and marks its data blocks as reusable. The actual bytes stay on disk untouched until the OS writes something else into that space. This is why recovery is possible at all — and why the less you use the drive after deleting something, the better your odds.

Metadata engine (metadata_recovery.py)

Walks the filesystem's own bookkeeping structures looking for entries flagged deleted-but-present (TSK_FS_META_FLAG_UNALLOC in Sleuth Kit terms). If the data blocks those entries point to are still intact, it reads them back in original size/order — correctly handling fragmentation, which carving cannot do. This is the accurate, "proper forensics" path, and it's tried first.

Carving engine (carver.py)

Ignores the filesystem entirely and scans raw bytes for known file signatures (FF D8 FF for JPEG, %PDF- for PDF, etc.), reading forward from each match to a known end-marker or a safety size cap. This is the fallback for cases where the filesystem index itself is gone (quick-formatted drive, corrupted table) — at the cost of no filenames, and occasional corruption if a file was fragmented across non-contiguous blocks.

It streams the source in 8 MB chunks with a small overlap window rather than loading the whole image into memory, so it scales to full-disk images, not just small USB sticks.

Limitations

• SSDs + TRIM. Modern SSDs (and most modern OSes) issue a TRIM command shortly after deletion that tells the drive to actually erase those blocks at the hardware level for wear-leveling. Once that's happened, no software — this tool, photorec, testdisk, anything — can recover the data. This mostly affects internal SSDs; most USB flash drives and SD cards do not implement TRIM, so recovery odds there are much better.
• Overwritten blocks produce partial/corrupt output, not an error — check file integrity after recovery.
• Carving fragmented files (a file split across non-contiguous disk regions) will produce corrupted output for formats without a reliable end marker. The metadata engine handles fragmentation correctly when it's available; carving is a last resort for exactly this reason.
• Encrypted volumes: if the disk was encrypted (LUKS, BitLocker, FileVault) and you don't have the key, recovered "data" is unreadable ciphertext — this tool doesn't and won't attempt to break encryption.

Comparison to existing tools

deep-recover is a learning/utility project, not a replacement for mature tools. For production forensic work also consider:

• photorec / testdisk — mature, battle-tested carving, more supported formats
• The Sleuth Kit / Autopsy — the full-featured version of what metadata_recovery.py does here, with a GUI and case management

Project layout

deep_recover/
  __init__.py
  signatures.py          # file signature database
  carver.py              # raw signature-carving engine
  metadata_recovery.py   # pytsk3-based filesystem-metadata engine
  cli.py                 # command-line entry point
requirements.txt
setup.py
README.md

License

MIT — see LICENSE.