Why does Extract API output extra bounding boxes and treat lines as rectangles?

[Rakesh Saini]

Hello Adobe Team,

I’m working with scanned invoices and using two APIs together:

• OCR API → to make the scanned PDF editable/searchable.

• Extract API → with parameters:

  const params = new ExtractPDFParams({ elementsToExtract: [ExtractElementType.TEXT, ExtractElementType.TABLES], addCharInfo: true });

This works, but I’ve noticed unexpected results when reviewing the JSON and trying to re-render the PDF:

1. The JSON output includes BBox attributes that add rectangular boxes around text and table elements.

2. When rendering from this JSON in Flutter, extra borders appear that do not exist in the original scanned PDF (e.g. double borders around tables, boxes around text).

3. It seems the API is treating every detected line or text area as a bounding rectangle, not just the actual drawn table/line borders from the original file.

   • Example: a single drawn line in the PDF becomes a rectangle in the JSON.

   • This makes it impossible to distinguish between real visual borders vs. bounding boxes used for OCR positioning.

[Joel Geraci]

You don't need the OCR step. Extract ignores the recognized text anyway. Extract API is able to perform OCR on image-only PDF automatically. But the output of Extract API was never designed to recreate the PDF. Also, in the real world, a line is a rectangle. It might be a very thin rectangle, but it's a rectangle. 

[Rakesh Saini]

Two Questions I have:
1. I have first applied the OCR API  then Used the Extracted API for that PDF. But into the extracted JSON why still getting  >> "is_scanned": true??

2. Does the extracted JSON contains complete information to recreate the original pdf throught any technology?

[Joel Geraci]

1) Because, as IU mentioned in my first response, Extract ignores the OCRed text. 

2) No.

[Rakesh Saini]

Thanks for your answer, is there any way to convert the Scanned pdf to Searchable/Editable using the ADOBE PDF API?

[Joel Geraci]

That's what the OCR API does.