Add support for ImageElements -> Parse images

[ic-xu]

The images in the PDF were lost after parsing before. Now RAG needs to use images, so I added the image extraction function.

[Filimoa]

Thanks for the PR - this looks awesome! I will look a take a deeper look into this asap

[ic-xu]

Okay, if you have any needs, please communicate in time. I will be happy to communicate with you.

[Filimoa]

PdfMiner looks good - I changed the image schema to encode the image data in base64 str to allow for easy serialization. We also added a mime type. Finally added a test - that looks good.

Looks like the pymupdf implementation is failing for me.

import openparse

basic_doc_path = "/Users/sergey/Downloads/pdf-with-image.pdf"
pdf_obj = openparse.Pdf(basic_doc_path)
parsed_basic_doc = ingest(pdf_obj)

Returns

ValidationError: 1 validation error for Bbox
  Value error, y1 must be greater than y0 [type=value_error, input_value={'x0': 72.0, 'y0': 719.54...0, 'page_height': 792.0}, input_type=dict]
    For further information visit https://errors.pydantic.dev/2.9/v/value_error

I noticed you flipped the coordinates? Maybe why?

fy0 = page.rect.height - node["bbox"][1]
fy1 = page.rect.height - node["bbox"][3]

To be honest, I'm not even sure how important it is to have this implemented for pymupdf because those documents are already OCRd and not sure how that affects this?

[Filimoa]

You can see the changes in the "parse-images-pdf-miner" branch - I can't seem to figure out how to merge it into this

[ic-xu]

PdfMiner looks good - I changed the image schema to encode the image data in base64 str to allow for easy serialization. We also added a mime type. Finally added a test - that looks good.

Looks like the pymupdf implementation is failing for me.

import openparse

basic_doc_path = "/Users/sergey/Downloads/pdf-with-image.pdf"
pdf_obj = openparse.Pdf(basic_doc_path)
parsed_basic_doc = ingest(pdf_obj)

Returns

ValidationError: 1 validation error for Bbox
  Value error, y1 must be greater than y0 [type=value_error, input_value={'x0': 72.0, 'y0': 719.54...0, 'page_height': 792.0}, input_type=dict]
    For further information visit https://errors.pydantic.dev/2.9/v/value_error

I noticed you flipped the coordinates? Maybe why?

fy0 = page.rect.height - node["bbox"][1]
fy1 = page.rect.height - node["bbox"][3]

To be honest, I'm not even sure how important it is to have this implemented for pymupdf because those documents are already OCRd and not sure how that affects this?

Indeed, for pymupdf, if the image has already been OCRed, then there is no need to parse the image anymore. However, I thought you would process the OCR in the process later, and only parse text in the text.

[ic-xu]

How about this, I'll submit the image parsing part for pdfminer first. As for pymupdf, I'll keep it as it is, without adding an image parsing module. Is that okay?

[Filimoa]

Sounds good!