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 notifications@github.com 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.
3 participants
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.