# US 12,306,882 B2 · Document matching using artificial intelligence

> Agents: this data is available as JSON from the patentgrep API and MCP server — see https://patentgrep.com/llms.txt

- Number: US 12,306,882 B2
- Document ID: US12306882B2
- Status: Granted
- Kind: B2
- Application number: 18/506,695
- Filed: 2023-11-10
- Published: 2024-03-14
- Granted: 2025-05-20
- Earliest claimed: 2021-03-30
- Assignees: SurePrep, LLC
- Applicants: Sureprep, LLC
- Inventors: David A. Wyle; Alexander James Sadovsky; William W. Hosek
- Web page: https://patentgrep.com/patent/US12306882B2

## CPC

- G06F 16/93: Document management systems
- G06N 3/0464: Convolutional networks [CNN, ConvNet]
- G06N 3/08: Learning methods
- G06N 3/09: Supervised learning
- G06N 5/046: Forward inferencing; Production systems
- G06V 30/1448: {based on markings or identifiers characterising the document or the area}
- G06V 30/412: Layout analysis of documents structured with printed lines or input boxes, e.g. business forms or tables
- G06V 30/414: Extracting the geometrical structure, e.g. layout tree; Block segmentation, e.g. bounding boxes for graphics or text
- G06V 30/416: Extracting the logical structure, e.g. chapters, sections or page numbers; Identifying elements of the document, e.g. authors
- G06V 30/418: Document matching, e.g. of document images
- G06N 3/045: Combinations of networks
- G06F: ELECTRIC DIGITAL DATA PROCESSING (computer systems based on specific computational models G06N)
- G06N: COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06V: IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING

## Abstract

The system is configured to create a generalized document automation framework that captures relevant data from documents based upon replicating historical human actions associated with a document. The system may use machine vision and natural language processing to match a new document to a document that was already human extracted in an existing corpus. This is accomplished by comparing both visual elements and textual elements. This match can be verified by statistical approaches by comparing the match metrics across multiple documents. After the match has been found and verified, the system then uses the historical extractions from the historical document and maps the extractions to similar regions in the new document based upon again both visual and text commonalities between documents. Data is then extracted from these regions of interest in the new document, sanity checked for data integrity against historical data, and then passed downstream for processing.

## Claims

### Claim 1

A method comprising:
storing, by a processor, cached metadata about a plurality of historical documents in a relational database to increase computation speed;
determining, by the processor, a mean square error (MSE) based on a difference between grayscale pixel intensities of each corresponding pixel of each image in a first content in one or more of a plurality of historical documents and a second content of a new document;
determining, by the processor, match metrics based on a percentage of the second content of the new document that matches the first content in one or more of the plurality of historical documents based on the cached metadata and;
extracting, by the processor, the second content from regions of interest in the new document based on the match metrics; and
automatically preparing, by the processor, documents using the second content.

### Claim 2 (depends on 1)

The method of claim 1, wherein the determining the match metrics further comprises:
comparing, by the processor, the second content of the new document to first content in one or more of the plurality of historical documents;
verifying, by the processor, the matching by comparing the match metrics across the plurality of historical documents; and
checking, by the processor, the integrity of the second content in the new document against the first content in the one or more of a plurality of historical documents.

### Claim 3 (depends on 1)

The method of claim 1, further comprising creating, by the processor, the match metrics for each of the one or more of a plurality of historical documents.

### Claim 4 (depends on 1)

The method of claim 1, further comprising transmitting, by the processor, the second content to a document processing system, wherein the document processing system automatically further prepares the documents using the second content.

### Claim 5 (depends on 1)

The method of claim 1, further comprising creating, by the processor, the cached metadata about the plurality of historical documents including regions of interest of objects, target data types, bag of words, identifier schema and known key-value pairs.

### Claim 6 (depends on 1)

The method of claim 1, further comprising the determining the match metrics based on the percentage of the second content of the new document that matches the first content in one or more of the plurality of historical documents based on at least one of a structural similarity index measure (SSIM) or an image similarity metric as determined by a deep neural network.

### Claim 7 (depends on 1)

The method of claim 1, wherein the determining includes comparing visual elements, textual elements, logos, fonts, lines, layouts, text from headers, text from values and text from tables.

### Claim 8 (depends on 1)

The method of claim 1, wherein the determining comprises:
decomposing, by the processor, the first content and the second content into features; and
predicting, by the processor using machine learning, a similarity probability based on the features.

### Claim 9 (depends on 1)

The method of claim 1, wherein the match metrics include a percentage overlap which comprises a number of words that are shared in the first content and the second content, divided by a total number of words between the first content and the second content.

### Claim 10 (depends on 1)

The method of claim 1, further comprising determining, by the processor, a subset of the plurality of historical documents that includes a subset of the first content that matches a subset of the second content to reduce processing time and conserve processing resources.

### Claim 11 (depends on 1)

The method of claim 1, wherein the determining is accomplished by location un-aware OCR detected text and visual document similarity indexing.

### Claim 12 (depends on 1)

The method of claim 1, wherein the determining further comprises detecting, by the processor, at least one of raw text of the second content in the new document or tables of the second content in the new document.

### Claim 13 (depends on 1)

The method of claim 1, wherein the determining further comprises pulling, by the processor, values, headers and text values from raw text of the second content in the new document, based on at least one of proximity, formatting or headers text values from raw text of the first content in the historical document.

### Claim 14 (depends on 1)

The method of claim 1, wherein the determining further comprises at least one of categorizing the second content similar to the first content or detecting tables of the first content in the historical document.

### Claim 15 (depends on 1)

The method of claim 1, wherein the determining further comprises pulling, by the processor, at least one of headers or text values from raw text of the first content in the historical document, based on at least one of proximity or formatting.

### Claim 16 (depends on 1)

The method of claim 1, wherein the determining further comprises comparing, by the processor, at least one of text headers, text values, X-Y coordinate positioning or visual elements between the new document and the historical document.
