unshred
unshred copied to clipboard
Write feature detector to determine angle of base line of text on a shred.
In addition to line detection we would like to determine angle of text baseline on shreds. This can be a valuable feature on it's own, because angle of a baseline is constant across the page and it might differ from 0 degrees because document was printed/shredded under slight angle. It also can determine position of each line on a shred (so we can use this to match it to another shred), number of lines, presence of text, etc.
Thanks to @fednep we have a solution that should be polished a little bit. Areas of improvement are:
- Adaptive binarisation similar to one that @xa4a ported in terms of #10
- Better accuracy
- Heuristics to suggest tags like has text, etc
- It should also return positions of text lines in relation to shred side
- Ideally: some way to evaluate algo using ground truth dataset.
Is under development in #5