Image Processing System Augmentation Parameter Optimization
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing image processing systems for detecting recognition targets, such as polyps in endoscope images, face challenges due to insufficient training data that does not adequately represent the features of polyps in actual detection environments, leading to reduced detection accuracy.
Innovation Solution
The system generates an augmented image set through data augmentation based on a candidate augmentation parameter, calculates augmentation and reference feature statistics, and determines the augmentation parameter to ensure the feature of the recognition target in the augmented image set matches that in a reference image set acquired in similar environments, thereby improving detection accuracy.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Quantity of substance
If data augmentation is applied to increase the number of training images, then the quantity of training data is improved, but the accuracy of detection for recognition targets with specific features deteriorates when the training image set does not include sufficient images with those features
Solution Approach 1:
The patent applies parameter changes by adjusting image characteristics (brightness, contrast, saturation, blur程度) through data augmentation to transform training images into augmented images that resemble actual detection environment images. This allows the training data to capture diverse feature distributions without requiring additional real-world images, thereby improving detection accuracy for recognition targets with specific features while maintaining sufficient training data quantity.
2Device complexity
If conventional data augmentation methods are used without feature statistic comparison, then the processing complexity is reduced, but the adaptability of the trained model to real-world environments deteriorates
Solution Approach 1:
The patent implements feedback by calculating feature statistics (mean and variance) of recognition target features from both the training image set and actual detection environment images, then using this feedback to determine optimal augmentation parameters. This feedback mechanism ensures the augmented images accurately represent real-world conditions, significantly improving model adaptability while the automated statistical comparison keeps the processing complexity manageable through algorithmic efficiency.
3Quantity of substance
If the training image set is expanded without considering feature distribution, then the quantity of training data is improved, but the manufacturing precision of the trained model deteriorates
Solution Approach 1:
The patent applies local quality by focusing data augmentation on specific feature characteristics (brightness, contrast, saturation, blur) that are locally important for recognition target detection. Instead of uniformly transforming all images, the method selectively adjusts specific feature parameters based on the calculated feature statistics, ensuring that the augmented images preserve the local quality and characteristic features necessary for accurate detection while maintaining diverse training data quantity.
Data Source
AI summary
An image processing system includes a memory storing a training image set and a reference image set, and a processor including hardware. The processor is configured to: generate an augmented image set by applying data augmentation to images included in the training image set; and determine an augmentation parameter based on a similarity between an augmentation feature statistic and a reference feature statistic, the augmentation feature statistic being a statistic of a feature of a recognition target calculated based on the augmented image set, the reference feature statistic being a statistic of a feature of the recognition target calculated based on the reference image set.