Hashed Template OCR for Semi-Structured Document Extraction
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Traditional optical character recognition (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 requiring manual data entry, which is time-consuming and error-prone, especially on resource-constrained devices.
Innovation Solution
A method that generates a database of document templates by processing and hashing electronic images, identifying spatial information of text values, and using these templates to extract and recognize text values from documents, even on mobile devices with limited resources.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If traditional OCR techniques are used on semi-structured documents with varying layouts, then the system can process documents, but the extraction accuracy deteriorates due to layout variations
Solution Approach 1:
The document is segmented into multiple regions of interest (ROIs) based on template matching. The template database contains segmented representations of document layouts, and each document is divided into corresponding regions to extract data systematically despite overall layout variations.
Solution Approach 2:
The system changes the parameter of document representation by converting documents into hashed template formats. This transformation allows the system to recognize documents based on their structural parameters rather than exact pixel matching, enabling accurate extraction across varying layouts.
2Measurement precision
If manual data entry is used, then data extraction accuracy can be maintained, but the time consumption and error rate increase
Solution Approach 1:
The system performs self-service by automatically extracting data from documents using template matching and OCR. The computing device autonomously identifies document types, selects appropriate templates, extracts regions, and recognizes text without requiring manual intervention, thereby reducing both time and errors.
Solution Approach 2:
The system performs preliminary actions by pre-processing documents into standardized template representations before actual data extraction. This preliminary template creation and matching enables rapid, accurate extraction without manual intervention during the extraction process itself.
3Measurement precision
If comprehensive template matching is performed on all documents, then extraction accuracy improves, but the processing complexity and resource requirements increase
Solution Approach 1:
The system applies partial action by matching documents against a subset of relevant templates rather than all possible templates. The template database is organized to enable selective matching based on document type and characteristics, reducing processing complexity while maintaining high accuracy for the target document set.
Solution Approach 2:
The system performs preliminary template generation and hashing for documents during initial processing. This preliminary action creates reusable template representations that can be quickly matched against future documents, reducing the complexity of repeated processing while maintaining comprehensive matching capability.
Data Source
Figure 1
Figure 2
Figure 3
AI summary
Techniques are disclosed for performing optical character recognition (OCR) by identifying a template based on a hash of a document. One embodiment includes a method for identifying a template associated with an image. The method includes receiving a digital image, a portion of the image depicting a first document, and extracting the portion of the image. The method further includes scaling the portion of the image and generating a first hash from the scaled image. The method further includes comparing the first hash to a set of hashes, each corresponding to a template. The method further includes selecting a first template as corresponding to the first document based on comparing the first hash to the set of hashes and extracting one or more sections of the portion of the image based on the selected first template. The method further includes performing OCR on the extracted one or more sections.