Segmentation Model Training for Small-Object Image Measurement

Resolve Bottlenecks,
Find Innovative SolutionsGenerate Solutions

Solution Overview

Problem

Existing image processing methods for images with numerous small objects are prone to human subjectivity, leading to instability, inefficiency, and low accuracy in obtaining measurement indicators.

Innovation Solution

An image processing method involving attribute transformation and the use of a trained segmentation model to automatically segment regions and calculate measurement indicators, reducing human intervention and enhancing accuracy.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Measurement precision

If manual comparison with preset template images is used to obtain measurement indicators, then the process can be performed with existing simple tools, but the processing stability and efficiency are greatly reduced and accuracy is low due to human subjectivity

Engineering Contradiction:

Improveaccuracy of measurement indicatorVSAvoidprocessing efficiency

Core Design Contradiction:

Measurement precisionVSProductivity

Solution Approach 1:

The patent replaces the manual mechanical comparison process with an automated image processing system that uses attribute transformation and segmentation models. The system automatically compares the target image with preset template images through computational algorithms, eliminating human subjectivity while maintaining high accuracy in measurement indicator extraction and significantly improving processing efficiency

Inventive Principle:

Principle #28Mechanics substitution (Replace mechanical system)

Solution Approach 2:

The patent implements self-service through automated attribute transformation and segmentation processes. The system performs self-comparison between target images and template images using preset algorithms, automatically generating measurement indicators without requiring manual intervention. This self-automated process ensures consistent accuracy and high processing efficiency

Inventive Principle:

Principle #25Self-service

2Reliability

Engineering Contradiction:

Improveprocessing stabilityVSAvoidcomplexity of image processing system

Core Design Contradiction:

ReliabilityVSDevice complexity

Solution Approach 1:

The patent replaces the unstable manual mechanical comparison process with a stable automated computational system. The system uses consistent algorithms for attribute transformation and image comparison, eliminating human subjectivity and ensuring high processing stability and reliability across different measurements

Inventive Principle:

Principle #28Mechanics substitution (Replace mechanical system)

Solution Approach 2:

The patent transforms images into attribute space (changing parameters from raw pixel values to meaningful attributes) before comparison. This parameter transformation standardizes the input data, ensuring consistent and reliable processing results while maintaining system stability across various image conditions

Inventive Principle:

Principle #35Parameter changes

3Productivity

If attribute transformation and segmentation model are used to automatically process images, then processing stability and efficiency are improved and accuracy is enhanced, but the system complexity increases

Engineering Contradiction:

Improveprocessing efficiencyVSAvoidcomplexity of image processing system

Core Design Contradiction:

ProductivityVSDevice complexity

Solution Approach 1:

The patent segments the image processing task into distinct stages: attribute transformation, segmentation, and measurement indicator extraction. By dividing the complex processing into modular segments, the system achieves high efficiency and accuracy while making the overall complexity more manageable through structured organization

Inventive Principle:

Principle #1Segmentation

Solution Approach 2:

The patent performs attribute transformation as a preliminary action before segmentation and measurement. This preprocessing step converts raw images into standardized attribute representations, simplifying subsequent processing steps and improving overall system efficiency while reducing the complexity burden on later stages

Inventive Principle:

Principle #10Preliminary action

Data Source

PatentUS12548160B2Method for training image processing model

Publication Date: 2026.02.10 TENCENT TECHNOLOGY (SHENZHEN) CO LTD

AI summary

This disclosure relates to a model training method and apparatus and an image processing method and apparatus. The model training method includes: obtaining a first sample image and a first standard region proportion corresponding to a first object in the first sample image; obtaining a standard region segmentation result corresponding to the first sample image based on the first standard region proportion; and training a first initial segmentation model based on the first sample image and the standard region segmentation result, to obtain a first target segmentation model.