End-to-End Deep Learning OCR for Handwritten and Typewritten Characters

Resolve Bottlenecks,
Find Innovative Solutions
Generate Solutions

Solution Overview

Problem

Existing optical character recognition (OCR) systems for both typewritten and handwritten text are time-intensive and prone to errors due to sequential processing steps, which can lead to information loss and distortion, especially when recognizing handwritten characters.

Innovation Solution

An end-to-end neural network architecture that processes images of text in a single computational step, incorporating global context and reducing noise, capable of distinguishing between handwritten and typewritten text, and generating segmentation masks and bounding boxes for characters, words, and lines.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Measurement precision

If sequential processing steps are used in OCR systems, then detailed processing of each step can be performed, but the process becomes time-intensive and computationally expensive

Engineering Contradiction:
Improvecharacter recognition accuracyVSAvoidprocessing time
Core Design Contradiction:
Measurement precisionVSLoss of time

Solution Approach 1:

The patent combines multiple sequential OCR processing steps (de-skewing, de-speckling, binarization, line removal, layout analysis, and character recognition) into a single end-to-end deep learning model. This unified model processes the entire OCR pipeline in one computational pass, eliminating the time delays and computational overhead associated with sequential step-by-step processing while maintaining recognition accuracy through integrated feature extraction and classification.

Inventive Principle:
Principle #5Merging (Combining)

Solution Approach 2:

The end-to-end deep learning model enables continuous processing of OCR tasks without the interruptions and data transfers between discrete sequential steps. The model maintains continuous computational flow from input image to output text, eliminating the idle time and computational reset overhead that occur when transitioning between separate processing stages in traditional sequential OCR systems.

Inventive Principle:
Principle #20Continuity of useful action

2Measurement precision

If multiple sequential processing steps are applied in OCR, then comprehensive text analysis can be achieved, but information loss and distortion occur in intermediate steps

Engineering Contradiction:
Improvetext recognition accuracyVSAvoidinformation loss in intermediate steps
Core Design Contradiction:
Measurement precisionVSLoss of information

Solution Approach 1:

By merging all OCR processing operations into a single end-to-end deep learning model, the patent eliminates intermediate processing steps where information loss and distortion typically occur. The unified model processes the image through integrated feature extraction, line detection, and character recognition in one continuous computational flow, preserving original image information throughout the processing pipeline without the degradation that occurs when data passes through multiple separate transformation stages.

Inventive Principle:
Principle #5Merging (Combining)

3Ease of manufacture

If handwritten character recognition is performed through sequential intermediate steps, then processing can be structured, but errors occur that cannot be later corrected

Engineering Contradiction:
Improvestructured processingVSAvoidrecognition reliability
Core Design Contradiction:
Ease of manufactureVSReliability

Solution Approach 1:

The patent merges structured processing with improved reliability by implementing a unified end-to-end deep learning model that maintains the organizational benefits of structured processing while eliminating the error propagation problem. The model integrates line detection, character segmentation, and recognition in a single coherent framework where each component benefits from the contextual information provided by others, allowing the system to maintain structure while achieving higher reliability through joint optimization of all processing stages.

Inventive Principle:
Principle #5Merging (Combining)

4Measurement precision

If sequential OCR processing is used, then detailed analysis can be performed, but significant computational resources are consumed

Engineering Contradiction:
Improveanalysis precisionVSAvoidcomputational resource consumption
Core Design Contradiction:
Measurement precisionVSUse of energy by moving object

Solution Approach 1:

The patent combines multiple computationally intensive OCR operations into a single end-to-end deep learning model that processes all tasks in one unified computational pass. This integration eliminates the redundant computations and data preprocessing overhead that occur when separate sequential models are applied, reducing overall computational resource consumption while maintaining the detailed analysis capabilities through the model's integrated feature extraction and multi-stage processing within the unified architecture.

Inventive Principle:
Principle #5Merging (Combining)

Data Source

PatentEP3712812B1Recognizing typewritten and handwritten characters using end-to-end deep learning
Publication Date: 2025.07.02 SAP SE
  • EP3712812B1 patent drawingFigure 1
  • EP3712812B1 patent drawingFigure 2A
  • EP3712812B1 patent drawingFigure 2B

AI summary

Disclosed herein are system, method, and computer program product embodiments for optical character recognition using end-to-end deep learning. In an embodiment, an optical character recognition system may train a neural network to identify characters of pixel images, assign index values to the characters, and recognize different formatting of the characters, such as distinguishing between handwritten and typewritten characters. The neural network may also be trained to identify groups of characters and to generate bounding boxes to group these characters. The optical character recognition system may then analyze documents to identify character information based on the pixel data and produce segmentation masks, such as a type grid segmentation mask, and one or more bounding box masks. The optical character recognition system may supply these masks as an output or may combine the masks to generate a version of the received document having optically recognized characters.