Object Recognition Using Point-of-Interest Text Detection

Resolve Bottlenecks,
Find Innovative Solutions
Generate Solutions

Solution Overview

Problem

Existing optical character recognition (OCR) methods suffer from lower recognition efficiency due to the recognition of entire text information of an object, which increases computational workload and reduces accuracy.

Innovation Solution

An object recognition method that identifies and recognizes point-of-interest (POI) text information using a target text detection model, reducing the need to process entire text information, and involves training an initial text detection model with detection boxes to generate a target text detection model.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Measurement precision

If the entire text information of an object is recognized by the text detection model, then comprehensive text extraction is achieved, but recognition efficiency deteriorates due to increased computational workload

Engineering Contradiction:
Improvecomprehensive text extraction accuracyVSAvoidrecognition efficiency
Core Design Contradiction:
Measurement precisionVSProductivity

Solution Approach 1:

The patent segments the text detection task by introducing point-of-interest (POI) text detection that focuses on specific key text regions rather than detecting all text information. The detection model is trained to identify and extract only the most relevant text areas, dividing the comprehensive text extraction task into selective POI detection and subsequent detailed recognition, thereby improving efficiency while maintaining accuracy for critical information

Inventive Principle:
Principle #1Segmentation

Solution Approach 2:

The patent extracts and prioritizes point-of-interest text information from the entire text content. By using POI detection to identify and extract only the most important text regions first, the system avoids processing all text information uniformly, thus improving recognition efficiency while ensuring comprehensive text extraction accuracy is maintained for the extracted POI regions

Inventive Principle:
Principle #2Taking out (Extraction)

2Loss of information

If the entire text information of an object is processed, then complete text recognition is achieved, but computational workload increases reducing processing speed

Engineering Contradiction:
Improvetext information completenessVSAvoidprocessing time
Core Design Contradiction:
Loss of informationVSLoss of time

Solution Approach 1:

The patent performs preliminary POI text detection before full text recognition. By first identifying and extracting point-of-interest text regions, the system prepares a reduced set of text information that needs detailed processing, thereby reducing processing time while maintaining text information completeness for the extracted POI regions

Inventive Principle:
Principle #10Preliminary action

Solution Approach 2:

The patent applies partial action by detecting and recognizing only the necessary POI text information rather than processing all text information. This selective approach reduces computational workload and processing time while ensuring that the essential text information is completely captured and recognized

Inventive Principle:
Principle #16Partial or excessive action

Data Source

PatentEP3979129B1Object recognition method and apparatus, and electronic device and storage medium
Publication Date: 2025.09.17 BEIJING BAIDU NETCOM SCI & TECH CO LTD
  • EP3979129B1 patent drawingFigure 1~2
  • EP3979129B1 patent drawingFigure 3
  • EP3979129B1 patent drawingFigure 4

AI summary

Disclosed are an object recognition method and apparatus, and an electronic device and a storage medium, which relate to the field of artificial intelligence. The method comprises: collecting an object to be subjected to recognition (S101); according to a target text detection model corresponding to the object to be subjected to recognition, carrying out screening and recognition on full text information corresponding to the object to be subjected to recognition, so as to obtain point-of-interest text information therefrom (S102); and carrying out recognition on the point-of-interest text information according to a preset text recognition model (S103). A target text detection model obtains point-of-interest text information by means of carrying out screening and recognition on full text information, such that the recognition of full text information in the prior art is avoided, thus saving recognition time, and improving the recognition efficiency.