Mixed Handwritten Printed Document OCR Accuracy

Resolve Bottlenecks,
Find Innovative Solutions
Generate Solutions

Solution Overview

Problem

Existing character recognition technologies, such as OCR for handwritten characters, often incorrectly recognize alphabet characters as numbers and symbols, leading to erroneous extraction of item values from document images, especially in cases where both handwritten and printed characters are present.

Innovation Solution

An image processing system and method that separates handwritten and printed character areas, performs specialized character recognition for each, integrates results, and uses a penalty-based likelihood calculation to accurately extract the item value from document images, considering the characteristics of handwritten and printed characters.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Adaptability or versatility

If OCR for handwritten character is used to extract item values, then handwritten characters can be recognized, but alphabet characters are erroneously recognized as numbers and symbols

Engineering Contradiction:
Improverecognition capability for handwritten charactersVSAvoidcharacter recognition accuracy
Core Design Contradiction:
Adaptability or versatilityVSMeasurement precision

Solution Approach 1:

The patent segments the character recognition process by separating handwritten character areas from printed character areas in the document image. Different OCR processing methods are then applied to each segment: printed character OCR for printed areas and handwritten character OCR for handwritten areas. This segmentation resolves the contradiction by allowing specialized processing for each character type, thereby maintaining high recognition capability while improving accuracy and preventing erroneous recognition of alphabet characters as numbers.

Inventive Principle:
Principle #1Segmentation

2Productivity

If character recognition processing is performed on mixed handwritten and printed documents, then all character areas can be processed, but erroneous recognition occurs due to applying uniform recognition methods

Engineering Contradiction:
Improveprocessing coverageVSAvoidextraction reliability
Core Design Contradiction:
ProductivityVSReliability

Solution Approach 1:

The patent applies local quality by assigning different recognition characteristics to different regions of the document. Printed character areas are processed with printed character OCR which has high accuracy for printed text, while handwritten character areas are processed with handwritten character OCR which is specialized for handwritten input. This local differentiation maintains high productivity across the entire document while improving reliability by using the most appropriate recognition method for each local region, thereby preventing erroneous extraction of item values.

Inventive Principle:
Principle #3Local quality

3Extent of automation

If item value extraction is performed using recognized character strings, then extraction can be automated, but erroneously recognized characters lead to wrong item values

Engineering Contradiction:
Improveautomation levelVSAvoiditem value extraction accuracy
Core Design Contradiction:
Extent of automationVSMeasurement precision

Solution Approach 1:

The patent segments the document image into handwritten and printed character areas before performing OCR, and then performs item value extraction based on the separately processed recognition results. This segmentation ensures that alphabet characters in handwritten sections are not erroneously converted to numbers, thereby maintaining high extraction accuracy while preserving full automation. The system automatically identifies character types, applies appropriate OCR, and extracts item values without human intervention, resolving the contradiction between automation and accuracy.

Inventive Principle:
Principle #1Segmentation

Data Source

PatentEP4036871B1Image processing apparatus, image processing method, program and storage medium
Publication Date: 2024.01.03 CANON KK
  • EP4036871B1 patent drawingFigure 1A~1B
  • EP4036871B1 patent drawingFigure 2
  • EP4036871B1 patent drawingFigure 3A~3C

AI summary

Character recognition processing suitable to a handwritten character area and a printed character area among character areas in a scanned image of a document is performed (121, 122, 123). Next, character recognition results for the handwritten character area and character recognition results for the printed character area are integrated and a likelihood indicating a probability of being an extraction target is calculated for a candidate character string that is an extraction candidate among the integrated character recognition results and a character string that is the item value is determined (124, 125). Then, at the time of the determination, different evaluation indications are used in a case where a character originating from the handwritten character area is included in characters constituting the candidate character string and in a case where such a character is not included.