char bounding box

[hhslepicka]

Hi,

I am trying to fetch each char's bounding box and compare that against the result generated by Apache PDFBox.
My goal is to reach the cyan box surrounding the g and the A displayed below:

Using pdfplumber I was able to fetch the information but it seems like the box is being affected by the other chars on the same text line.

My code with pdfplumber:

import pdfplumber
file = "./teste_final.pdf"
pdf = pdfplumber.open(file)
p0 = pdf.pages[0]
p0_im = p0.to_image()
p0_im.reset()
p0_im.draw_rects([p0.chars[-13], p0.chars[-1]])

The code from Apache PDFBox which generated the first image is located here.

Here is my test PDF file:
https://drive.google.com/file/d/16V0BzcwP50f-iWhR-eRfGkZ5_o83BD6a/view?usp=sharing

Here is the full image from the first crop

:

Thank you for your help and developing pdfplumber.

[jsvine]

Very interesting test cases you have here, @hhslepicka, and thank you for sharing!

Practically speaking, this question would be better suited for the pdfminer.six repository; that's the dependency we use to extract character-location information. I'm unfortunately not deeply-familiar enough with the specifics of their character-bounding-box algorithm to be able to diagnose the situation myself.

But I do wonder: What bounding boxes do you get when you remove the g from the A a g line? Does the bounding box for A still descend below the baseline?

[hhslepicka]

Thank you, @jsvine.
I was wondering the same so I tried with a new PDF to try and answer this question.

Here is the debug with pdfplumber/pdfminer.six:

Here is the debug with PDFBox:

I will try to reach out to the pdfminer.six team as well. If I get any input I will link it here.

Thanks again for your help.

[jsvine]

Thanks for sharing that. That does seem to be the output I'd expect from pdfminer.six. And in the case of pdfplumber, I think those are indeed the bounding boxes we typically want — especially for the purpose of grouping characters on the same line. That said, I could definitely see it being useful if pdfminer.six provided users with the option for closer-cropped bounding boxes. Curious to hear what you learn!