Y-coordinates Swapped in Generated Box File, Other Minor Issues

[nikkelj]

Hi again,

So I seem to have solved my issues now (again, kudos for building a nice system). In order to get this working in my configuration/fonts, I had to do the following in multipage_tif:

1. subprocess.call was for some reason causing errors in running ImageMagick convert, so I switched the code to use os.system(cmd) instead (single string, not string array)

2. I changed _save_tif() to save as a 300DPI tif. This may be unnecessary in many cases, but most FAQ's/postings I see recomment at least this.
   tif.save(self.indiv_page_prefix + str(page_number) + '.tif',dpi=(300,300))

3. Most important: I do not know if it is something specific to my configuration that is causing this (I am using Tesseract 3.02, Python 2.7.5), but I had to swap char_y1 and char_y0 in the call to self._write_boxline inside _fill_pages() to get this to work. The y-coordinates were swapped opposite the defined convention otherwise.

4. The code in _fill_pages() also ignores cases where fonts have a baseline offset. To remedy this, I collected the offsets from the font definition using:
   offset= self.font.getoffset(char)

and then adjusted char_x0/x1 and char_y0/y1 by offset[0] and offset[1] respectively.

5. Depending on the metrics of the font chosen, you may also need to bias the newline y_position so that the lines do not overlap, and adjust the character-to-character spacings (adjust x_pos and y_pos to taste). May want to specify these as inputs or config file items so that folks know they may need to adjust them.

6. I am finding that font.getsize(char) does not always return accurate boundaries from given font files, particularly for some punctuation. So, I am adding a custom offset hash to manually offset specific characters in a given font.

Hope that helps,

-Jon

[brouberol]

Hi @nikkelj.

Thank you so much for your contributions.
Would you mind sending a proper pull request, so that I could integrate them into the master branch.
By doing that, you would basically enable 3.02 support :)

Cheers

[nikkelj]

Certainly, will do so tonight.

-Jon

On Tue, Oct 29, 2013 at 5:00 AM, Balthazar Rouberol <
[email protected]> wrote:

Hi @nikkelj https://github.com/nikkelj.

Thank you so much for your contributions.
Would you mind sending a proper pull request, so that I could integrate
them into the master branch.
By doing that, you would basically enable 3.02 support :)

Cheers

—
Reply to this email directly or view it on GitHubhttps://github.com//issues/9#issuecomment-27293712
.

Jonathan Nikkel
Mobile: 970-310-8710
Email: [email protected]

[tfmorris]

@nikkelj Ping! :-)

Sounds like this might actually be multiple pull requests, but sounds like 3 is the critical one for 3.02 support.

[nikkelj]

Sorry, forgot about this for a while, pull request is up now