Image Processing Model Selection via Score Difference Evaluation
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing pattern matching methods in image processing often result in erroneous detections due to the reliance on trial and error for model generation, leading to increased processing time and potential manufacturing defects, especially in industrial applications where high-speed processing is crucial.
Innovation Solution
An image processing method that generates multiple provisional models by extracting connected edges from a reference image, calculates matching scores against evaluation images, and determines a reliable model based on the difference between the highest and second-highest similarity scores to minimize erroneous detection.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If multiple models are used for pattern matching to reduce erroneous detection, then reliability is improved, but processing time increases
Solution Approach 1:
The patent performs model evaluation and selection in advance before actual pattern matching. Multiple candidate models are evaluated using evaluation images to calculate evaluation values, and the best model is selected beforehand. This preliminary action ensures high detection reliability without increasing processing time during actual manufacturing operations.
Solution Approach 2:
The system automatically evaluates and selects the optimal model without requiring manual intervention or trial-and-error by operators. The model selection is performed autonomously based on evaluation values calculated from evaluation images, making the system self-sufficient in optimizing detection reliability.
2Reliability
If manual model generation with detailed settings is performed to reduce erroneous detection, then detection reliability is improved, but ease of operation deteriorates
Solution Approach 1:
The system automatically performs model evaluation and selection without requiring operator skills or manual adjustment. The automatic evaluation process calculates evaluation values for multiple candidate models and selects the optimal one, eliminating the need for operators to have specialized knowledge or perform trial-and-error adjustments.
Solution Approach 2:
The patent changes the approach from manual parameter adjustment to automatic parameter optimization. Instead of operators manually setting model parameters based on empirical rules, the system automatically evaluates different models using calculated evaluation values and selects the best parameters objectively.
3Reliability
If model generation is performed through trial and error to reduce erroneous detection, then detection reliability is improved, but loss of time increases
Solution Approach 1:
The system performs comprehensive model evaluation in advance using evaluation images before actual manufacturing. Multiple candidate models are assessed and the optimal model is selected beforehand, eliminating the need for time-consuming trial-and-error adjustments during production operations.
Solution Approach 2:
The patent introduces an evaluation value calculation mechanism that provides objective feedback on model performance. By calculating evaluation values from evaluation images and using them to select the best model, the system replaces subjective trial-and-error with objective feedback-driven selection, significantly reducing model generation time.
Data Source
AI summary
An image processing apparatus includes a provisional model generation portion, a matching score obtaining portion, an evaluation portion, and a determination portion. The provisional model generation portion is configured to generate a plurality of provisional models. The matching score obtaining portion is configured to perform the pattern matching between each of the plurality of provisional models and each of a plurality of evaluation images, and obtain a first matching score group that is a set of first matching scores indicating a highest degree of similarity and a second matching score group. The evaluation portion is configured to calculate an evaluation value from the first matching score group and the second matching score group. The determination portion is configured to determine a matching model from among the plurality of provisional models on a basis of the calculated evaluation value.