Line detection #2
Conversation
* added sorting for fast clusterization
much better lines identification
* fixed line count
|
Are you happy with implementation and want me to merge or need more time/help to polish it? |
|
Константин, добрый вечер. Как с вами связаться, подскажите? |
|
Добрый вечер,
|
|
Добавил в скайпе. On Sun, Aug 24, 2014 at 10:36 PM, mr-const [email protected] wrote:
|
* better noise reduction (i hope)
* shred-dependent line length
| cv2.line(shred, (x1, y1), (x2, y2), (255, 255, 0, 0), 2) | ||
| cv2.imwrite('../debug/houghlines_%s.png' % name, shred) | ||
|
|
||
| params['Lines Count'] = len(lines[0]) |
There was a problem hiding this comment.
"Lines count" doesn't seem to be too useful for unshredding. What about detecting major lines direction? e.g. make a histogram with buckets 0,1º,..,89º and find most popular. The result might potentially be useful to group the shreds with similarly sloped lines.
Conflicts: unshred/split.py
Line detection PR #2 follow-up.
Запилил неплохо работающий детектор линий (т.е. полосок, которые на страницах).
Нужно ещё добавить фильтрацию линий, которые накладываются друг на друга. Добавить для удобства теггирование "вертикальные линии, горизонтальные, косые, вотэвер".
Ну и уменьшить число false positives.