Apryse Answers: Diving into PDF Data Extraction

PDF documents are designed to put document style and content together in a single self-contained file, keeping formatting and appearance consistent across experiences. However, PDF documents aren’t machine-friendly, especially when they originate as scans or images of text.

OCR is one tool that helps detect text, but it doesn’t preserve document structure or formatting. When a PDF with columns, headings, footers, and tables is processed by OCR, the output is a wall of text, straight across the page. This jumbles content from different table columns, discards important structure such as headings, and results in more work to make the output usable.

Apryse Answers is our video series where Developer Relations Manager April Schuppel finds real developer questions on Reddit, and answers them using Apryse SDKs. In episode 4, April tackles questions about PDF Data Extraction.

Find the code from the video on Github.

Apryse Smart Data Extraction is a set of SDK tools that solve the challenges of PDF data extraction. Smart Data Extraction goes beyond OCR to capture valuable data from a range of document structures, including tables, headings, footers, and unstructured text such as paragraphs.

Try the Demo in our Showcase

Smart Data Extraction includes the following primary modes of intelligent extraction:

Let’s take a look at how April answered three questions from Reddit, then dive into the documentation for building Extraction workflows.

This user needs to programmatically read text from a PDF. This could be required to help route documents at intake for processing or storage, or to use document data to power other automation, such as AI summaries, document generation, or trigger events.

Apryse SDK can go beyond OCR to improve outcomes and reduce errors for these use cases. For example, Apryse Document Classification detects document types, and key value extraction pulls essential data from unstructured text, such as names and numbers.

I had a patron who had a document that she needed to edit. She did not have the original Word file. Another colleague thought we could scan her doc to PDF and then extract the text from that PDF to insert into a new Word doc that she could edit. But I cannot find if this is a thing that can be done. Anyone know?

Scanning a document results in an image of text, which requires OCR to convert into searchable text. In addition to Smart Data Extraction to power downstream processing of scanned documents, Apryse PDF to DOCX conversion could serve this user.

For my work, I need to extract the text from PDFs quite a lot and also keep the formatting. I used to do it manually, but recently found pdftotext by xpdf, which speeds the process up. However, this only creates a .txt file with plain text and no formatting (only bold, italics, underlined, and regular would be enough).

Is there a tool which extracts the text from a PDF and keeps formatting?

Losing the formatting of a document is a major reason why developers seek out more capable data extraction solutions beyond OCR.

To try any Apryse SDK in your environment, start with your free trial key.

The Smart Data Extraction module is part of the Apryse Server SDK. To install the Server SDK, follow the instructions in our documentation based on your specific framework and language requirements.

The Server SDK supports a number of frameworks, runtimes, and languages in all major platforms for delivering applications from a single codebase.

When using Python on Windows or Linux you can install the package via PIP with this command:

You can find other installation methods, such as npm for node.js or installing the package directly, in the documentation guide.

For error handling purposes, it is generally advisable to test whether the module is available via the IsModuleAvailable function. Since the Data Extraction suite consists of multiple modules, an extra parameter is used to clarify the component to test.

The simplest way to use Document Classification, Document Structure Recognition, and the other tools in Smart Data Extraction, is to specify the name of the input PDF file and the name of the output JSON file, then select the required engine:

The engines are:

This code outputs document data in a JSON file, but additional configuration is possible for each module. Check out the options, such as outputting a JSON string instead, in the documentation.

Before extraction begins, documents often need to be cleaned, normalized, or digitized. Apryse supports a full preprocessing toolkit—so your inputs are structured, accurate, and AI-ready.

These capabilities are modular and can be used independently or together, depending on your workflow:

These preprocessing tools improve downstream performance across:

No hallucinations. No unstructured text blobs. Just labeled, model-ready JSON.

Sample code shows how to use the Apryse Data Extraction module to extract tabular data, document structure and form fields from PDF documents. Find the sample here in other languages.

If you have any questions about your Smart Data Extraction trial and how to get started with licensing for your use case, contact sales. Check out the Github for Apryse Answers to find the specific project demonstrated in the video!

Apryse Answers continues next week with questions about annotations and document collaboration. See you there!