Image Processing Sequence Generation Using Probability-Based Learning Set Selection
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing image processing techniques require excessive time to generate a desired image processing sequence, especially when dealing with multiple types of defect-free images, due to the limitations in the number and types of inspection area images that can be processed.
Innovation Solution
A method that selects a smaller number of learning sets and iteratively refines the image processing sequence by comparing output images with target images, using a probability-based selection of learning sets to increase the efficiency of the sequence generation process, until a predetermined reference condition is met.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If the number of learning images used for evaluation is increased to restrain overlearning, then the evaluation accuracy is improved, but the processing time increases significantly
Solution Approach 1:
The patent pre-processes learning images to extract feature values before the genetic programming evaluation phase. By preparing feature values in advance, the system reduces the computational burden during evaluation, allowing for more learning images to be used without proportionally increasing processing time.
Solution Approach 2:
The patent divides the learning images into multiple groups and processes them in stages. Instead of evaluating all learning images simultaneously, the system segments the evaluation process to manage computational complexity and reduce overall processing time while maintaining evaluation accuracy.
2Measurement precision
If multiple types of good-quality images without defect are prepared for evaluation, then the evaluation accuracy is improved, but the processing time increases due to the limited number and types of inspection area images
Solution Approach 1:
The patent creates synthetic or augmented versions of good-quality images by applying various transformations and compositions. Instead of requiring multiple physically different defect-free images, the system generates multiple variants through copying and transforming existing high-quality images, thereby expanding the evaluation dataset without proportionally increasing processing time.
Solution Approach 2:
The patent applies parameter changes to existing good-quality images, such as adjusting brightness, contrast, or other image parameters, to generate multiple types of inspection area images. This approach allows the system to create diverse evaluation images from a limited set of source images, improving evaluation accuracy while maintaining processing efficiency.
Data Source
AI summary
In a method for generating an image processing sequence, when selecting a learning set, a probability of selection of the learning set having a predetermined value, of a plurality of subject evaluation values corresponding respectively to two or more learning sets calculated in a previous routine, is increased.