OCR on Semi-Structured Documents Using Spatial Classifier Segmentation

Resolve Bottlenecks,
Find Innovative Solutions
Generate Solutions

Solution Overview

Problem

Traditional OCR techniques fail to efficiently extract data from semi-structured documents like tax forms due to variations in layout and design, leading to sub-par results and the need for manual data entry, which is time-consuming and error-prone, especially on resource-constrained devices like mobile phones.

Innovation Solution

A classifier model is trained using spatial attributes to identify label regions in documents without performing OCR, allowing for the accurate distinction between labels and values, and subsequent OCR is performed only on value regions to extract corresponding values, reducing the need for manual data entry and improving efficiency on resource-constrained devices.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Extent of automation

If traditional OCR techniques are used on semi-structured documents with varying layouts, then the system can process documents without manual intervention, but the extraction accuracy deteriorates due to layout variations

Engineering Contradiction:
Improveautomation of data extractionVSAvoiddata extraction accuracy
Core Design Contradiction:
Extent of automationVSMeasurement precision

Solution Approach 1:

The patent segments the document processing task into distinct phases: first identifying label regions using spatial attributes and trained classifier models, then performing OCR only on value regions. This segmentation allows the system to handle layout variations by treating labels and values differently, improving extraction accuracy while maintaining automation.

Inventive Principle:
Principle #1Segmentation

Solution Approach 2:

The patent performs preliminary classification of document regions into labels and values before executing OCR. By using spatial attributes and trained classifier models to identify label regions first, the system prepares the document structure in advance, enabling more accurate subsequent OCR on value regions only.

Inventive Principle:
Principle #10Preliminary action

2Loss of information

If OCR is performed on entire documents to ensure all data is captured, then completeness of data extraction is improved, but processing time and computational resources increase

Engineering Contradiction:
Improvecompleteness of data extractionVSAvoiddata processing time
Core Design Contradiction:
Loss of informationVSLoss of time

Solution Approach 1:

The patent extracts and identifies label regions from documents using spatial attributes and classifier models, then performs OCR only on the remaining value regions. This extraction approach ensures all necessary data is captured while significantly reducing the area subjected to computationally intensive OCR processing, thereby decreasing processing time.

Inventive Principle:
Principle #2Taking out (Extraction)

Solution Approach 2:

Instead of applying OCR to the entire document (excessive action), the patent applies OCR only to value regions after identifying and excluding label regions (partial action). This partial application of OCR maintains data completeness while reducing computational overhead and processing time.

Inventive Principle:
Principle #16Partial or excessive action

3Loss of information

If OCR is performed on entire documents to capture all data, then data extraction completeness is improved, but computational resource consumption increases

Engineering Contradiction:
Improvecompleteness of data extractionVSAvoidcomputational resource consumption
Core Design Contradiction:
Loss of informationVSUse of energy by moving object

Solution Approach 1:

The patent extracts label regions from documents using spatial attributes and trained classifier models, then performs OCR only on value regions. This extraction strategy ensures complete data capture while minimizing the computational resources required for OCR processing by limiting it to necessary regions only.

Inventive Principle:
Principle #2Taking out (Extraction)

Solution Approach 2:

The patent applies OCR partially only to value regions rather than excessively to the entire document. This partial application maintains information completeness while significantly reducing computational resource consumption, making the system more efficient on resource-constrained devices.

Inventive Principle:
Principle #16Partial or excessive action

4Measurement precision

If manual data entry is used to handle layout variations, then data extraction accuracy is improved, but labor time and error rate increase

Engineering Contradiction:
Improvedata extraction accuracyVSAvoidmanual data entry time
Core Design Contradiction:
Measurement precisionVSLoss of time

Solution Approach 1:

The patent enables the system to automatically handle layout variations through self-service mechanisms: trained classifier models automatically identify label regions using spatial attributes, and OCR automatically extracts values from identified regions. This automation eliminates the need for manual intervention while maintaining high extraction accuracy, thereby reducing both time and error rate.

Inventive Principle:
Principle #25Self-service

Data Source

PatentEP3437019B1Optical character recognition in structured documents
Publication Date: 2020.09.30 INTUIT INC
  • EP3437019B1 patent drawingFigure 1
  • EP3437019B1 patent drawingFigure 2
  • EP3437019B1 patent drawingFigure 3

AI summary

Techniques are disclosed for facilitating optical character recognition (OCR) by identifying one or more regions in an electronic document to perform the OCR. For example a method for identifying information in an electronic document includes obtaining a set of training documents for each template of a plurality of templates for the electronic document, extracting spatial attributes for at least a first label region and at least a first corresponding value region from the set, and training a classifier model based on the extracted spatial attributes, wherein the classifier model is used to identify the information in the electronic document. The spatial attributes represent a position of at least the first label region and at least the first value region within the electronic document.