Image Text Recognition Using Semantic and Layout Fusion

Resolve Bottlenecks,
Find Innovative Solutions
Generate Solutions

Solution Overview

Problem

Current methods for recognizing text in images fail to achieve precise recognition results due to variations in accuracy among different recognition methods, leading to inconsistent and inaccurate output.

Innovation Solution

A method that combines multiple recognition results using semantic and feature information of the image, including named entity and layout information, through an attention mechanism-based neural network to determine the most accurate recognition result.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Reliability

If multiple recognition methods are used to improve text recognition accuracy, then the reliability of recognition results is improved, but the device complexity increases due to needing to integrate and evaluate multiple recognition systems

Engineering Contradiction:
Improvetext recognition accuracyVSAvoidsystem complexity
Core Design Contradiction:
ReliabilityVSDevice complexity

Solution Approach 1:

The patent combines multiple text recognition methods into a unified system that processes images through different recognition algorithms simultaneously. The system merges the results from these multiple methods and uses evaluation mechanisms to select the most accurate recognition result, thereby improving overall text recognition accuracy while managing system complexity through integrated processing.

Inventive Principle:
Principle #5Merging (Combining)

Solution Approach 2:

The patent introduces an intermediary evaluation mechanism that assesses the quality and accuracy of recognition results from multiple methods. This intermediary layer analyzes semantic information, layout consistency, and other features to determine which recognition result is most reliable, acting as a mediator between multiple recognition methods and the final output.

Inventive Principle:
Principle #24Intermediary (Mediator)

2Measurement precision

If semantic information and feature information are combined to determine the most accurate recognition result, then the measurement precision of text recognition is improved, but the loss of time increases due to additional processing steps

Engineering Contradiction:
Improverecognition result accuracyVSAvoidprocessing time
Core Design Contradiction:
Measurement precisionVSLoss of time

Solution Approach 1:

The patent performs preliminary extraction of semantic information and feature information from recognition results before final evaluation. By pre-processing and organizing this information in advance, the system reduces the computational burden during the final selection process, thereby improving recognition accuracy while minimizing additional processing time.

Inventive Principle:
Principle #10Preliminary action

Solution Approach 2:

The patent applies different evaluation criteria to different aspects of recognition results. Semantic information is used to evaluate the meaningfulness of recognized text, while feature information (such as layout and positioning) is used to verify spatial consistency. This localized evaluation approach improves overall precision without requiring uniform processing of all information types.

Inventive Principle:
Principle #3Local quality

Data Source

PatentEP4027267B1Method and system for identifying text in image
Publication Date: 2026.02.25 HUAWEI TECH CO LTD
  • EP4027267B1 patent drawingFigure 1
  • EP4027267B1 patent drawingFigure 2
  • EP4027267B1 patent drawingFigure 3~4

AI summary

A method, an apparatus, and a system for recognizing a text in an image are provided. The method includes: obtaining a plurality of recognition results of a to-be-recognized text in an image according to a plurality of recognition methods (S201); obtaining semantic information of the recognition results (S202); obtaining feature information of the image, where the feature information of the image can represent information around the to-be-recognized text in the image (S203); and determining a target recognition result of the to-be-recognized text from the plurality of recognition results based on the feature information of the image and the semantic information of the plurality of recognition results (S204). According to the method, a recognition result in the recognition results that includes the feature information of the image can be recognized because the feature information that is of the image and that can represent the information around the to-be-recognized text in the image is combined when a most accurate recognition result is determined from the plurality of recognition results. Therefore, accuracy of determining the most accurate recognition result from the plurality of recognition results can be improved, that is, a precise recognition result can be obtained.