Charged Particle Inspection Image Normalization for Stable ANN Processing
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Charged particle microscopy images are sensitive to imaging conditions, leading to degradation in analysis and manipulation results, especially when data is noisy or of poor quality, making it challenging for Artificial Neural Networks (ANNs) and Convolutional Neural Networks (CNNs) to maintain stability and accuracy without retraining.
Innovation Solution
A method that adjusts image quality parameters such as resolution, noise, and contrast to match a moderate quality standard, allowing ANNs and CNNs to process images consistently without the need for retraining, by either enhancing or deteriorating images based on their initial quality to align with the network's training conditions.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If ANNs and CNNs are used for image processing in production environments, then analysis and manipulation results can be obtained, but the quality degrades when imaging conditions vary (noisy data, poor focus)
Solution Approach 1:
The patent applies preliminary action by pre-processing images to match the quality characteristics of training data before analysis. The system determines current image quality parameters, compares them with training data characteristics, and applies quality adjustment processing to transform the image into characteristics matching the training set. This preliminary transformation ensures that images processed in production environments have the same quality characteristics as training images, eliminating degradation caused by varying imaging conditions.
2Reliability
If retraining is performed to account for new imaging conditions, then analysis quality can be maintained, but computation time and cost increase significantly
Solution Approach 1:
The patent applies copying by creating a transformed copy of the input image that matches training data characteristics, rather than retraining the model. Instead of retraining the ANN/CNN with new data under different imaging conditions, the system copies the image and applies quality adjustment processing to create a version that replicates the training conditions. This approach maintains analysis quality without the computational overhead of retraining.
3Measurement precision
If image quality is improved to maintain consistent processing, then analysis accuracy increases, but the complexity of image modification increases
Solution Approach 1:
The patent applies feedback by implementing a closed-loop system that determines current image quality parameters, compares them with training data characteristics, and applies appropriate quality adjustment processing. The system includes a feedback mechanism that continuously monitors image quality parameters and adjusts the processing accordingly. This feedback-based approach achieves consistent image quality transformation while managing complexity through automated parameter determination and comparison.
Data Source
AI summary
The invention relates to a method implemented by a data processing apparatus, comprising the steps of receiving an image; providing a set-point for a desired image quality parameter of said image; and processing said image using an image analysis technique for determining a current image quality parameter of said image. In the method, the current image quality parameter is compared with said desired set-point. Based on said comparison, a modified image is generated by using an image modification technique. The generating comprises the steps of improving said image in terms of said image quality parameter in case said current image quality parameter is lower than said set-point; and deteriorating said image in terms of said image quality parameter in case said current image quality parameter exceeds said set-point. The modified image is then output.