OCR solution for scanning scanned documents

Question

We want to scan scanned documents (PDF/Images), and extract the meaningful information as key value pair or any format. like When a suppliers sends us invoice as scanned pdf, our solution should extract and look for the meaningful header and line level details.

We tried open source and commercial solutions like ABBYY, Amazon, Tesseract, and Google OCR, but the accuracy is not up to the mark and also the output is just plain text in case of open source solutions.

Kindly suggest any best open source or commercial solutions (OCR) which best fits for the above requirements.

Answer 1

You could utilize the LEADTOOLS Invoice Recognition SDK, creating a master form would allow you to identify the exact pieces of the invoice that you need and get the results of a scanned image as a key-value pair. As a disclaimer this is a paid SDK and I work for this vendor. The below link additionally has a sample project of how this could be done in Java.

https://www.leadtools.com/support/forum/posts/t12833-HOW-TO--Use-AutoForms-in-Java

// Setup
codecs = new RasterCodecs();
ocrEngine = OcrEngineManager.createEngine(OcrEngineType.LEAD);
ocrEngine.startup(codecs, null, null, null);
repository = new DiskMasterFormsRepository(codecs,
                 "C:\\Users\\Public\\Documents\\LEADTOOLS Images\\Forms\\MasterForm Sets\\Invoice");
autoEngine = new AutoFormsEngine(repository, ocrEngine, null);

// Prepare the form image with some preprocessing (optional)
unknownForm = codecs.load(unknownFormFilepath);
//prepareForm(unknownForm);

// Attempt to match the unknownForm to one of the masterforms in repository
AutoFormsRunResult runResult = autoEngine.run(unknownForm, null, null, null);
if (runResult != null) {
         // Print out the form information on success
         printOutValues(runResult);
}

Answer 2

You are really looking for data extraction, not so much OCR (although it does USE OCR). Some cloud options include AWS Textract and Azure Cognitive Services You are looking for Forms Recognition, not just OCR. AWS Textract and Azure Forms Recognizer have functionality of this type

Answer 3

Global Search by square 9 would be my suggestion. https://www.square-9.com/products/enterprise-content-management/

It allows for infinite scanning options, to categorize documents based on OCR info, lets you target regions of pages to extract information from form fields on common formatted documents, can scan read barcodes, and allows for elaborate workflows to be built like split and send part of the document to this archive, part to that archive, etc.. Essentially scan, read, make decisions, and then store however you want. OnPrem, and client is zero footprint (browser based)

Abiut the most robust I have worked with in my carrer.