Xerox 2.0 Bug in this script

[tobltobs]

The script makes some funny things with for e.g. an "8":

[jasonlfunk]

Yeah. I know the script doesn't work very well. I'm not actively supporting it.

[RusEu]

same problem here :(

[RusEu]

Also with B letter ..

[RusEu]

On commenting this:
# if count_children(index, h_, contour) > 2:
# if DEBUG:
# print "\t skipping, is a container of letters"
# return True
I get the B and the 8 , but very dark and the OCR doesn't recognize it well. Any idea how to solve this?

[varunkumar2310]

max edge-boxes that can occur inside a text edge-box = 2
but each edge-box has 2 contours : outer contour + inner contour
take case of A : total contours = outer contour of A + inner contour of A + outer contour of delta + inner contour of delta = 4 contours
And an edge-box can had max of 2 edge-boxes completely enclosed in it.
(max enclosing edge-boxes for English language = 2)
Which means max contours that can be enclosed inside a text ( inside outer contour of text )
= inner contour of text + 2 * max edge-boxes = 1 + 2 * 2 = 5
change the max contours depending on language if necessary

Therefore in include_box:
modify the lines to :
if is_child(index, h_) and count_children(get_parent(index, h_), h_, contour) <= 5:
if count_children(index, h_, contour) > 5:

I have coded in OpenCV 3.1, C++ , and the following change works for me.
Correct me if my logic is wrong.

Images:
original

processed_with_5

rejected_with_5

processed_with_2

rejected_with_2

[guddulrk]

I am getting following error:

inf = file(scratch_text_name_root + '.txt')
NameError: name 'file' is not defined

Any help please??

[tsjason]

@guddulrk No idea. It looks like you might be missing some libraries or something. But I haven't used this in ages so I'm not sure.

[itcthienkhiem]

same error.

[itcthienkhiem]

B still problem ?
How can fix it ?

[NightFury13]

max edge-boxes that can occur inside a text edge-box = 2
but each edge-box has 2 contours : outer contour + inner contour
take case of A : total contours = outer contour of A + inner contour of A + outer contour of delta + inner contour of delta = 4 contours
And an edge-box can had max of 2 edge-boxes completely enclosed in it.
(max enclosing edge-boxes for English language = 2)
Which means max contours that can be enclosed inside a text ( inside outer contour of text )
= inner contour of text + 2 * max edge-boxes = 1 + 2 * 2 = 5
change the max contours depending on language if necessary

Therefore in include_box:
modify the lines to :
if is_child(index, h_) and count_children(get_parent(index, h_), h_, contour) <= 5:
if count_children(index, h_, contour) > 5:

I have coded in OpenCV 3.1, C++ , and the following change works for me.
Correct me if my logic is wrong.

Images:
original

processed_with_5

rejected_with_5

processed_with_2

rejected_with_2

Pappu! Thanks dude! :D @varunkumar2310

[sstefanov]

Because of used Canny each contour is recognized twice.
This is the reason for not proper handling B and 8 chars. Children are 4, not 2.

To fix this just replace 2 with 4 in:
if is_child(index, h_) and count_children(get_parent(index, h_), h_, contour) <= 2:
must be
if is_child(index, h_) and count_children(get_parent(index, h_), h_, contour) <= 4:
and
if count_children(index, h_, contour) > 2:
must be
if count_children(index, h_, contour) > 4:

See my pull request.