Veo Palm Camera PDB → JPEG Decoder

A reverse-engineered decoder for the .pdb picture files of the Veo Photo Traveler for Palm Handhelds, a small SD-card camera for Palm OS devices from the early 2000s.

The camera stored every picture as a Palm Database (.pdb) with a proprietary compression ("Type 4"). The Windows software that converted them to JPEG is long gone. This decoder makes the pictures readable again.

The full story with pictures: palm2000.com

Quality

Compared with a JPEG created by the original Veo software:

• PSNR: 24.46 dB
• Correlation: 0.984
• Speed: about 0.4 seconds per picture (640×480)

The result is not bit-identical (the original software uses its own demosaicing and JPEG settings), but visually there is no difference.

Usage

Requirements: Python 3.8+, numpy and Pillow.

pip install numpy Pillow

Single file:

python3 src/veo_pdb_decoder.py picture.pdb [output.jpg]

Whole folder:

python3 src/veo_pdb_decoder.py --batch path/to/pdbs/ path/to/output/

With the embedded thumbnail:

python3 src/veo_pdb_decoder.py picture.pdb --thumb

Two sample files are included in data/:

python3 src/veo_pdb_decoder.py data/3840520404.pdb iphone.jpg

How it works

• Container: Palm Database with a 78-byte header. Record 0 holds 25 bytes of metadata, the last record a 36×28 pixel RGB565 thumbnail, and the records in between the compressed image, four rows per record (120 records → 480 rows).
• Raw Bayer data: The camera stores raw sensor data in a GBRG pattern, not a processed picture:

  Row 0 (even): Gb  B  Gb  B  Gb  B  ...
  Row 1 (odd):  R   Gr R   Gr R   Gr ...
  

• Type 4 compression: Each record is its own bitstream (MSB first). Each call decodes two rows in three byte-aligned passes:
  1. Row 0, odd columns (blue)
  2. Row 1, even columns (red)
  3. Row 0 even and row 1 odd columns (green), alternating, predicted from the other row
• Codes per value: 1 = repeat the previous value · 0 00000 + 8 bits = literal · 0 SDDDD = difference ±1…15.
• Picture: gray-world white balance, then bilinear demosaicing to 640×480.

Typical white balance gains: R 1.0×, Gr 1.6×, Gb 1.7×, B 4.8× (depends on the light).

The format was reverse-engineered from the Veo HotSync conduit for Windows (COMConduit.dll). The whole way there, including the dead ends, is documented in docs/REVERSE_ENGINEERING.md; a short technical overview is in docs/OVERVIEW.md.

Files

Path	Content
src/veo_pdb_decoder.py	The decoder (command line tool)
src/veo_decoder_v3.py … v9*.py	Earlier versions, kept to document the process
src/veo_* (other)	Analysis and debugging tools used during reverse engineering
data/	Two sample .pdb files
docs/	Overview and reverse engineering notes

License

MIT, see LICENSE.