How to extract text based on parts from a PDF file in JSON format?

Question

From this file https://github.com/pymupdf/PyMuPDF-Utilities/blob/master/text-extraction/Dart.pdf

I would like to get this kind of result:

{
    "file": {
        "title": "Dart Programming Language Specification",
        "1 Scope": {
            "text": "This Ecma standard specifies the syntax and semantics of the Dart program-ming language. It does not specify the APIs of the Dart libraries except where those library elements are essential to the correct functioning of the language itself (e.g, the existence of class Object with methods such as noSuchlethod, runtimeType."
        },
        "2 Conformance": {
            "text": "A conforming implementation of the Dart programming language must pro-vide and support all the APIs (libraries, types, functions, getters, setters, whether top-level, static, instance or local) mandated in this specification. A conforming implementation is permitted to provide additional APIs, but not additional syntax, except for experimental features in support of null-aware cascades that are likelv to be introduced in the next revision of this specification."
        },
        "3 Normative References": [
            {
                "text": "The following referenced documents are indispensable for the application of this document. For dated references, only the edition cited applies. undated references, the latest edition of the referenced document (including any amendments) applies.",
                "1": "The Unicode Standard, Version 5.0, as amended by      Unicode 5.1.0, or successor.",
                "2": "Dart API Reference, https://api.dartlang.org/"
            }
        ]
        ...
    }
}

My first idea was to perform layout detection with Deep Learning using OCR techniques (notably Tesseract) with Detectron2 and libraries such as deepdoctection and layout-parser. But after some tests, the detection of the text layout doesn't seem to be taken into account. I can only extract "the overall layout" with titles, text boxes and tables. The sorting of the detection according to the coordinates of the boxes must be done afterwards.
My second idea is to first convert the PDF file into a text file with a text extraction approach preserving the layout. Several solutions exist like PyMuPDF. And then perform a processing on the text file that generates the outline with the parts (titles, subtitles, texts etc..) as a dictionary according to the white spaces of the file. But this solution seems not robust because I can have pdf files where there is no indentation on the titles of the parts.

Is there a way to perform layout detection of each part of the layout with its text and these associated subparts?

How to extract text based on parts from a PDF file in JSON format?

Answers (1)

Related Questions