[MRG] Add saturation threshold option for low contrast tables

[NoReflex]

I found out this problem while trying to parse a table with low contrast background color. The -back option didn't work for low contrast areas such as the last row. So, I've added a new option -color (--process_color_background) which increases the contrast to guarantee accurate table parsing.

Here's camelot (master) result:

Here's my branch with -color option enabled:

As you can see, we add another step which is basically a binary threshold for low saturation vs no saturation.
Now the borders are way more pronounced and camelot has no issue detecting all the rows.

[codecov-io]

Codecov Report

Merging #203 into master will decrease coverage by 0.60%.
The diff coverage is 21.42%.

@@            Coverage Diff             @@
##           master     #203      +/-   ##
==========================================
- Coverage   88.26%   87.65%   -0.61%     
==========================================
  Files          14       14              
  Lines        1542     1555      +13     
  Branches      350      351       +1     
==========================================
+ Hits         1361     1363       +2     
- Misses        127      137      +10     
- Partials       54       55       +1     

Impacted Files	Coverage Δ
camelot/io.py	100.00% <ø> (ø)
camelot/utils.py	81.26% <ø> (ø)
camelot/image_processing.py	82.14% <8.33%> (-12.38%)	⬇️
camelot/cli.py	86.77% <100.00%> (+0.11%)	⬆️
camelot/parsers/lattice.py	94.14% <100.00%> (+0.03%)	⬆️

---

Continue to review full report at Codecov.

Legend - Click here to learn more
Δ = absolute <relative> (impact), ø = not affected, ? = missing data
Powered by Codecov. Last update d17dc43...9161ef3. Read the comment docs.

[vinayak-mehta]

@NoReflex Thanks for the PR! The results look great! I don't have enough image processing and opencv background so will have to read up on the cv2.COLOR_BGR2HSV option. I also have a question, do you think this new code could also handle other background line cases? That way we could just add it as an enhancement to the earlier option instead of creating a new one.

[NoReflex]

@vinayak-mehta The cv2.COLOR_BGR2HSV is just a colorspace transformation from RGB to HSV (Hue, Saturation, Value).
As for the question, this would fail if the table's cell colors are gray/colorless. That's why it's an option.
Technically it's still an enhancement, because the -color flag can only be used with -back, but yeah, I get your point and I think it's better if it's a separate option for handling edge cases.

EDIT: By failing, I just mean that the result will be worse than using the plain option, the code is pretty bulletproof, it's just some simple numpy array transformations.

[MartinThoma]

Hey!

As camelot is dead, we try to build a maintained fork at pypdf_table_extraction.

Do you want to open the PR against that branch so that we can merge your improvement?