Made a change to 1D decoding which looks for 100% instead of 150% of the start and end patterns when searching for quiet zones, but in exchange for that reduction, we reject rows whose quiet zones would run off either edge of the image.
The result is complete elimination of all false positives in the new batch of 40 images I just checked in. Bettter than that, we come up with a net gain of 38 new barcodes decoded (counting rotation and try harder variations). We even get some of the reflective images that were impossible up to this point.
git-svn-id: http://zxing.googlecode.com/svn/trunk@620
59b500cc-1b3d-0410-9834-
0bbf25fbcc57