Automated methods and systems for retrieving information from scanned documents
Inventors
Ward, Michael • Westerman, Dax • Reeves, Ruth • Wrenn, Jesse
Assignees
Interested in licensing this patent?
MTEC can help explore whether this patent might be available for licensing for your application.
Publication Number
US-12530538-B2
Publication Date
2026-01-20
Expiration Date
Abstract
Automated systems and methods for retrieving information from scanned documents are described herein. An example method includes reading a scanned document; and extracting a page image from the scanned document, where the page image includes text and structured data. The method also includes executing a structured data detection algorithm to detect the structured data contained in the page image and identify a plurality of elements of the structured data. Additionally, the method includes performing optical character recognition (OCR) to convert the text contained in the page image to first text data; and performing OCR to convert respective contents of each of the plurality of elements of the structured data to second text data. The method further includes executing a natural language processing (NLP) algorithm to retrieve information from the first and second text data.
Core Innovation
The invention describes an OCR+structured-data detection+NLP system that converts a scanned document into computable text and extracts information. A method reads a scanned document and extracts a page image that includes text and structured data. OCR converts the text in the page image into first text data.
The method executes a structured data detection algorithm to detect the structured data and identify a plurality of elements. The structured data detection algorithm includes performing cell boundary detection by transforming the page image to a black and white image, detecting a white background using a pixel flood process, and detecting one or more lines bounding the plurality of cells. The method then performs table boundary detection to identify one or more tables comprising the plurality of cells, and performs table structure detection to identify rows and columns within each table.
The method performs OCR again to convert respective contents of each structured-data element into second text data. It then executes a natural language processing (NLP) algorithm to retrieve information from the first and second text data. The extracted information is based on both OCR text from the page image and OCR text from the structured-data elements.
Claims Coverage
The independent claims are directed to a computer-implemented method and a system, covering OCR of scanned page text, structured data detection for cells, tables, rows, and columns using specified image-processing steps, OCR of structured elements, and NLP-based information retrieval. The core claim set includes five inventive features.
Reading and extracting a page image with text and structured data
Reading a scanned document and extracting a page image from the scanned document, wherein the page image includes text and structured data.
OCR to first text data from page image text
Performing optical character recognition (OCR) to convert the text contained in the page image to first text data.
Cell boundary detection via black-and-white transformation, pixel flood background, and bounding lines
Executing structured data detection algorithm steps that perform cell boundary detection to identify a plurality of cells within the structured data by transforming the page image to a black and white image; detecting a white background within the structured data using a pixel flood process; and detecting one or more lines bounding the plurality of cells.
Table boundary detection and table structure detection to rows and columns
Performing table boundary detection to identify one or more tables within the structured data, wherein each table comprises one or more of the plurality of cells; and performing table structure detection to identify a plurality of rows and columns within each table.
OCR to second text data from structured-data elements
Performing OCR to convert respective contents of each of the plurality of elements of the structured data to second text data.
NLP retrieval from first and second text data
Executing a natural language processing (NLP) algorithm to retrieve information from the first and second text data.
System implementation of the OCR+structured detection+NLP pipeline
A system comprising a processor and memory with computer-executable instructions configured to read a scanned document, extract a page image with text and structured data, convert page text to first text data using an OCR module, detect structured data using a structured data detection module, perform table boundary detection and table structure detection to identify rows and columns, convert structured elements to second text data using the OCR module, and retrieve information using an NLP module from the first and second text data.
Across the independent claims, the inventive coverage is centered on OCR from page image text, structured data detection that identifies cells and tables using black-and-white transformation, pixel flood background detection, and bounding lines, table structure detection to rows and columns, OCR of structured elements to second text data, and NLP-based information retrieval from both OCR outputs.
Stated Advantages
Documented Applications
No documented applications found
Interested in licensing this patent?