Material Recognition Using Texture Segmentation

Resolve Bottlenecks,
Find Innovative Solutions
Generate Solutions

Solution Overview

Problem

Existing material recognition systems face challenges in accurately identifying materials independent of object shape, leading to variations in recognition accuracy due to shape attributes, and often rely on limited and skewed training data.

Innovation Solution

A method and apparatus that utilize a neural network to generate and train material models by extracting texture regions from object images, excluding shape attributes and focusing on material characteristics, thereby creating high-quality training data that improves recognition performance and reduces shape-related accuracy variations.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Adaptability or versatility

If material recognition systems use object images containing shape attributes, then they can recognize materials in various object contexts, but recognition accuracy varies due to shape-related interference

Engineering Contradiction:
Improvematerial recognition in various object contextsVSAvoidrecognition accuracy
Core Design Contradiction:
Adaptability or versatilityVSMeasurement precision

Solution Approach 1:

The patent segments object images into texture regions and shape regions, then uses only texture regions for material recognition training. This segmentation isolates material-specific texture information from shape-related interference, allowing the system to recognize materials across various object contexts while maintaining consistent accuracy by excluding shape attributes that cause recognition variations.

Inventive Principle:
Principle #1Segmentation

2Adaptability or versatility

If training data includes diverse object shapes, then the system can handle various materials, but shape attributes create skewed and limited training data quality

Engineering Contradiction:
Improvehandling various materialsVSAvoidtraining data quality
Core Design Contradiction:
Adaptability or versatilityVSReliability

Solution Approach 1:

The patent extracts only texture regions from object images, removing shape attributes entirely from the training data. This extraction process creates high-quality training data that focuses exclusively on material characteristics, improving reliability by eliminating shape-related skewness while maintaining the ability to handle various materials through diverse texture samples.

Inventive Principle:
Principle #2Taking out (Extraction)

3Loss of information

If the system processes complete object images, then it captures all visual information, but shape attributes interfere with pure material characteristic extraction

Engineering Contradiction:
Improvevisual information captureVSAvoidmaterial characteristic extraction accuracy
Core Design Contradiction:
Loss of informationVSMeasurement precision

Solution Approach 1:

The patent segments object images to separate texture regions containing material characteristics from shape regions. By processing only the segmented texture regions, the system prevents shape attributes from interfering with material characteristic extraction while preserving all relevant visual information about material properties through focused texture analysis.

Inventive Principle:
Principle #1Segmentation

Data Source

PatentEP3324339B1Method and apparatus to perform material recognition and training for material recognition
Publication Date: 2024.07.03 SAMSUNG ELECTRONICS CO LTD
  • EP3324339B1 patent drawingFigure 1
  • EP3324339B1 patent drawingFigure 2
  • EP3324339B1 patent drawingFigure 3

AI summary

Provided are method and apparatuses related to material recognition and training. A training apparatus for material recognition generates training data associated with a material by generating a texture image having a texture attribute from an object image and recognizing material information from the texture image using a material model.