Continuing the spirit of video documentation, Glen helped me record a short demo (1min 41sec):

I used the same Exemplar SVM classification technique as I was using in the LA exhibition. I had trouble using a logitech webcam due to auto-exposure, but the ps3-eye worked fine. (Though tweaking exposure is still, well, tweaking.)

Unlike with the LA project, this does not require a precise alignment of the cards. Instead, I find the four corners of the card (using heuristics from a threshold-contour) and apply a perspective unwarp (I do expect them to be face up, plus or minus 45 degrees). Also unlike the LA project, I trained the classifier on the perfect digital versions of the business cards (i.e. the PDF that was printed), rather than camera images:

That saved a significant amount of time, and it seems to work just as well.

The computer vision code can recognize several cards simultaneously, but  I didn’t figure out what I would do with that sort of input data:

Finally, at Glen’s suggestion, I used Applescript/iBooks for the computer re-integration. The script is simple—it just hits Cmd+f and types in the text from the card—but I don’t know how long iBooks is going to take before I can hit “down, return” and its search is sometimes quite slow, so I wait three seconds, which is annoying:

I will bring it to Twitter Book Club now… let’s see if it can be part of a conversation, not its subject.

-R.M.O.