Image Processing for Defect Attribute Detection Using Actual Pixel Size
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing image processing methods for defect detection in structures like concrete surfaces fail to accurately determine defect attributes due to insufficient consideration of actual image size, leading to poor determination accuracy when learning data similarities are low.
Innovation Solution
An image processing apparatus and method that utilizes actual size information per pixel to select and correct defect attributes using machine learning models, ensuring accurate determination by aligning image and model size ratios.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Ease of operation
If a model is selected based only on image capturing position and angle similarity, then model selection is simplified, but determination accuracy deteriorates when actual size information is not considered
Solution Approach 1:
The patent extends the model selection criteria by adding actual size information per pixel as a new parameter. The selection unit now compares multiple parameters including image capturing position, angle, and crucially, actual size information. This parameter expansion resolves the contradiction by maintaining ease of automated selection while improving determination accuracy through more comprehensive similarity matching.
Solution Approach 2:
The patent adds a new dimension to the model selection process by incorporating actual size information per pixel. Instead of selecting models based solely on 2D image characteristics (position and angle), the system now operates in a 3D parameter space that includes scale/size information. This additional dimension enables more accurate model matching while maintaining automated selection efficiency.
2Measurement precision
If multiple models with different learning data contents are prepared to cover various images, then determination accuracy for different image types is improved, but device complexity increases
Solution Approach 1:
The patent implements a feedback mechanism where the selection unit automatically chooses the most appropriate model based on comparing actual size information and other parameters between the input image and learning data. This feedback-driven selection eliminates the need for manual model management and reduces device complexity while maintaining high determination accuracy across various image types. The system adapts to different image conditions automatically through parameter-based model selection.
Solution Approach 2:
The patent introduces dynamic model selection based on the actual size information and characteristics of the input image. Instead of using a fixed set of models for different image types, the system dynamically selects the most appropriate model by comparing parameters in real-time. This dynamic approach reduces the number of models needed while maintaining high accuracy across varying image conditions.
3Measurement precision
If learning data with different actual sizes are used to create multiple models, then determination accuracy for various scales is improved, but the complexity of model management and selection increases
Solution Approach 1:
The patent addresses scale variation by incorporating actual size information per pixel as a selection parameter. Instead of creating separate models for different scales, the system selects the appropriate model by comparing the actual size parameter between the input image and stored learning data. This parameter-based selection approach maintains high determination accuracy across various scales while significantly reducing model management complexity.
4Device complexity
If a single model is used for all images, then device complexity is reduced, but determination accuracy deteriorates when image characteristics vary significantly
Solution Approach 1:
The patent implements dynamic model selection that adapts to varying image characteristics. The selection unit compares actual size information and other parameters between the input image and learning data to automatically choose the most appropriate model. This dynamic selection mechanism maintains device simplicity with a unified model management system while achieving high determination accuracy across diverse image types through parameter-based model matching.
Data Source
AI summary
An image processing apparatus comprises an obtainment unit configured to obtain first actual size information of an image including a defect, and a determination unit configured to determine an attribute of the defect included in the image using the first actual size information and a model generated by learning in advance.