enhancement: allow setting image block crop padding parameter

[christinestraub]

Closes #2320 .

Summary

In certain circumstances, adjusting the image block crop padding can improve image block extraction by preventing extracted image blocks from being clipped.

Testing

• PDF: LM339-D_2-2.pdf
• Set two environment variables EXTRACT_IMAGE_BLOCK_CROP_HORIZONTAL_PAD and EXTRACT_IMAGE_BLOCK_CROP_VERTICAL_PAD
  (e.g. EXTRACT_IMAGE_BLOCK_CROP_HORIZONTAL_PAD = 40, EXTRACT_IMAGE_BLOCK_CROP_VERTICAL_PAD = 20

elements = partition_pdf(
    filename="LM339-D_2-2.pdf",
    extract_image_block_types=["image"],
)

[Coniferish]

@christine
This is what I'm using to test this, but I'm not seeing a difference in the image that's saved from the main branch and this one:

from unstructured.partition.pdf import partition_pdf
filename = "example-docs/LM339-D_2-2.pdf"
elements = partition_pdf(
    filename=filename,
    extract_image_block_types=["image"],
    strategy="hi_res",
    extract_image_block_output_dir="output/"
)

Do I need to define some additional param to increase the padding?

[Coniferish]

Do the .rst files in docs/source/ need to be updates with these changes?

[christinestraub]

@christine This is what I'm using to test this, but I'm not seeing a difference in the image that's saved from the main branch and this one:

from unstructured.partition.pdf import partition_pdf
filename = "example-docs/LM339-D_2-2.pdf"
elements = partition_pdf(
    filename=filename,
    extract_image_block_types=["image"],
    strategy="hi_res",
    extract_image_block_output_dir="output/"
)

Do I need to define some additional param to increase the padding?

@Coniferish To see a difference in the image, you'll need to set two environment variables EXTRACT_IMAGE_BLOCK_CROP_HORIZONTAL_PAD and EXTRACT_IMAGE_BLOCK_CROP_VERTICAL_PAD.
I updated testing instruction in description.

[christinestraub]

Do the .rst files in docs/source/ need to be updates with these changes?

Updated the partition.rst file in docs/source/ :) Thanks!

[Coniferish]

LGTM once we get that docker test passing :)

Want to check something before approval

[Coniferish]

@christinestraub, one last question: I want to make sure that this should NOT update the element.text that's detected even if the image now includes the full text (in this example "Output" is now visible, but it's still not a part of element.text)

[christinestraub]

@christinestraub, one last question: I want to make sure that this should NOT update the element.text that's detected even if the image now includes the full text (in this example "Output" is now visible, but it's still not a part of element.text)

I didn't expect to update the element.txt. @cragwolfe What do you think of this? Do we need to update element.text as well when extracting by adding padding to the detected image block?

[cragwolfe]

@christinestraub, one last question: I want to make sure that this should NOT update the element.text that's detected even if the image now includes the full text (in this example "Output" is now visible, but it's still not a part of element.text)

I didn't expect to update the element.txt. @cragwolfe What do you think of this? Do we need to update element.text as well when extracting by adding padding to the detected image block?

I think the primary intent of this PR is with respect to the image artifacts saved, so do not need to update the .text in this PR. It's a good callout though and worth tracking in another github issue. But, there is a fair bit of nuance -- it seems find to extend OCR .text, but also want to be careful not to duplicate other text if the bbox overlaps another element. Or, don't allow a bbox overlap in the first place for OCR, only extend to min(EXTRACT_IMAGE_BLOCK_CROP_HORIZONTAL_PAD,distance_to_next_bbox).

[cragwolfe]

LGTM!