Wine Labler

Describes a prototype App that scans a wine bottle on a moving turnstile so as to produce a flattened scan of its labels. This involved finding the bottle edges, then grabbing a small slice in the center 30 times a second. These slices where then algorithmically joined to produce the label.

The presentation shown at Lot18 was the basis of a later blog post residing on the Lot18 web site [a link to it appears in the final slide]

Transcript

1. In Search of Wine Labels The Story of WineLabler
2. The Beginning Sam -> David: “We should let users save bottle photos for later viewing.” David -> Sam: “Too hard to view on iPhone— lets add a label photo for this purpose.” Sam -> David: “We don’t get them from suppliers—no way to force them to provide.”
3. There has to be a way… A few bottles use water-based glue, can soak them off (but fewer and fewer) Boiling Water + Razor + Safety Glasses + Gloves Baking / Really Sticky Plastic Sheets / Hair Dryer / Ammonia / … Tried a hand-held scanner but no go.
4. Cold Water Soak 24 Hours Later...->
5. Boiling Water
   
6. Water Soluble Is Best
7. iPhone 4
8. iPhone App Objectives System oenophiles could personally construct and use with reproducible results Lot18 could use to produce label only images Total Equipment Cost < $100 and easy to get Require iPhone 4 or newer Port to the Mac easily
9. Scanner/Stitcher Needs A video camera that can deliver HD images at a high rate (10-20fps) Turntable of appropriate speed Computer that can grab the images More processing power is better!
10. Affordable Environment +
11. WineLabeler
12. Architecture Focus and Calibrate the iPhone camera Using one snapshot, delineate the bottle parts: shoulder, vertical sides, label bottom With turntable spinning, grab tiny slivers and stitch them together, matching newest to previous Save images, provide for viewing, deleting, and uploading
13. Bottle Delineation and Stripe Target
14. Building the Image Current turntable rotates once every 33 seconds Begin with a label area having significant variation for rotational speed detection. Scan for 50 seconds, grabbing 500 images. Match current to previous, then blend into final image.
15. The First Image
16. Where to Go Professional
    - Mac + external HD camera
    - iPad III Amateur
    - iPhone
    - iPad 3
17. This presentation was the genesis of an article wrote for the Lot18 blog:
 
 http:/ /blog.lot18.com/2012/07/01-a-new-gadget