How to set background color opacity for div block in insert_htmlbox? It seems blank.

[nissansz]

How to set background color opacity for div block in insert_htmlbox? It seems blank.

text = """

some text

and some text
with a different background

and some more text

"""

[nissansz]

Now, textwriter morph effect seems perfect.

[nissansz]

textwriter seems faster than insert_htmlbox

[JorjMcKie]

It sure is! A lot.
We must not forget what happens in background: HTML/CSS parsing, invocation of the HarfBuzz library for text shaping.
The method has a certain area of purpose and is excellent there.
But if used beyond that it will show limits.

[nissansz]

Smaller font size, better effect.

[JorjMcKie]

Very nice!

[JorjMcKie]

See here:

import fitz


rect = fitz.Rect(100, 100, 200, 150)

text = """<div style="background-color:yellow;">Text 1</div>
<div style="background-color:rgb(0,255,0);
text-align:center;color:red;">Text 2</div>"""

doc = fitz.Document()

page = doc.new_page()
page.insert_htmlbox(rect, text)
page.draw_rect(rect, color=(1, 0, 0))
doc.ez_save(__file__.replace(".py", ".pdf"))

So this does not fully fill the rectangle with background color.

There is a way to completely exploit the rectangle's area, by using a crazily large font size like css="* {font-size: 50px}". In that case, the method automatically scales down by an optimized factor and gives you this:

[nissansz]

It seems TextWriter is more powerful for text effect.

[JorjMcKie]

As I said, your use case is not the usual one, and definitely very different from the intention of insert_htmlbox.

[nissansz]

Understood. Maybe one day, when javascript is supported in insert_htmlbox, it will be very powerful.

[JorjMcKie]

What you need here is the opposite: detail influence on the layout. the htmlbox intention is to take away this consideration.

[nissansz]

You are pretty correct and expert.

[JorjMcKie]

Understood. Maybe one day, when javascript is supported in insert_htmlbox, it will be very powerful.

It is already very powerful - think of reproduction of images, and HTML tables, and above all correct output of languages requiring text shaping linke Devanagari, Bengali, Tamil, etc.

Textwriter is incapable of doing this.

[nissansz]

Will try tables in future, still need time to train table models to recover table structure. Any good method to recover table structure from pdf image?

[JorjMcKie]

Not really, just OCR, then page.find_tables().
Problem is that OCR mostly cannot recognize grid lines, which breaks usual table recognition.
So you are dependent on analyzing text positions to sort of guess what might be a table.

[JorjMcKie]

AWS Textract does a good job, but depends on Internet connection, is an AI machinery ... and slow.

[nissansz]

Thank you. Will train my own models to test.