User’s Guide

Data for the 3DFV database was collected through various campaings in Switzerland, at different locations. Each of the campaigns and outcomes are summarized next:

  • Foire du Valais (FDV): 89 subjects provided data from both index fingers in a single data-acquisition session with a single attempt. For each subject finger, 5 snapshots were taken. The unique subject identifiers vary between 1 and 94, but numeration is not contiguous.

  • Idiap (IDI): 50 subjects provided data from both left and right index, middle and ring fingers in 3 data acquisition sessions, each with 2 attempts. Two more subjects provided data for the same fingers, but for only 1 acquisition session (subjects 131 and 147). The unique subject identifiers range from 101 to 153. Numeration is contiguous.

Filename Structure

Filenames inside the 3D Fingervein are structured like this:

<database-root>/<client:%03d>/<session:%d>/<attempt:%d>/<client:%03d>-<age:%02d>-<gender:%s><skin:%s><occ:%s><side:%s><finger:%s><session:%d><attempt:%d><snap:%d><cam:%d>.png

Each field can have these values:

  • client: integer > 0

  • age = integer > 0

  • gender = str, ‘m’ or ‘f’

  • skin (color) = str, ‘1’..’6’ or ‘x’

  • occ(upation) = str, ‘0’..’9’ or ‘x’

  • side = str, ‘l’ or ‘r’

  • finger = str, ‘t’, ‘i’, ‘m’, ‘r’, ‘l’

  • session = int > 0

  • attempt = int > 0

  • snap = int > 0

  • cam = int, one of 1, 2, 3 or S (1 is left, 2 is central, 3 is right and S means the “stitched” image using the three cameras)

Protocols

Central

The central protocol only makes use of images from the central camera of the prototype (camera 2). Each finger is supposed to be considered a different individual. Existing samples are divided like this:

  • Training set: All subjects with only one session from FDV and Idiap

  • Development set: Snapshots (2 per finger) and 2 attempts using the system from all subjects with 3 sessions from Idiap. Modelling uses snapshots from session 1, probing uses snapshots from sessions 2 and 3. While probing, one should match every file listed as probe against all models in the set. In this protocol, 1 snapshot image generates one model for the subject.

Stitched

The stitched protocol is the same as central, but uses the stitched images for training and development.