Image Anomaly Detection With Twin-Pixel Relationships
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing nondestructive testing techniques struggle to accurately detect anomalies in objects without causing damage, particularly in quality-control processes where defects need to be identified efficiently.
Innovation Solution
A method and system using specially-configured computing devices for anomaly detection in images, which involves obtaining training images, selecting associated pixels, calculating value relationships, and detecting anomalies based on twin-pixel differences and MAD scores, eliminating the need for direct pixel-value comparison and reducing systematic bias.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If direct pixel-value comparison is used for anomaly detection, then the detection process is simple, but the accuracy is reduced due to systematic bias and sensitivity to texture variations
Solution Approach 1:
The patent introduces an intermediary representation (e.g., gradient maps, frequency domain transforms, or learned feature representations) between the raw pixel values and the anomaly detection process. This intermediary transforms the data in a way that removes systematic bias and reduces sensitivity to texture variations, thereby improving detection accuracy without requiring direct pixel-value comparison
Solution Approach 2:
The patent transforms the pixel data from the original value domain to a different parameter domain (such as gradient domain, frequency domain, or other feature spaces). This parameter transformation changes the characteristics of the data to eliminate systematic biases and make the anomaly detection more robust to texture variations while maintaining a manageable process complexity
2Measurement precision
If pixel-value adjustments and perfect image alignment are required, then the detection accuracy may improve, but the ease of operation deteriorates due to additional preprocessing requirements
Solution Approach 1:
The patent designs the detection method to be self-correcting or self-aligning, where the algorithm inherently compensates for misalignment and texture variations without requiring manual preprocessing. The method uses local comparisons or relative measurements that are invariant to global transformations, allowing the system to operate effectively on raw images without extensive adjustment
Solution Approach 2:
Instead of trying to make images perfectly aligned and adjusted before detection, the patent inverts the approach by designing detectors that are insensitive to misalignment and texture variations. The method detects anomalies based on local patterns and relationships that remain consistent even when global image properties vary, eliminating the need for rigorous preprocessing
3Reliability
If traditional nondestructive testing techniques are used, then the process is straightforward, but the reliability of anomaly detection is insufficient for quality-control applications
Solution Approach 1:
The patent extends the detection process from simple pixel-value comparison to multi-dimensional analysis, incorporating spatial relationships, gradient information, frequency characteristics, or other dimensional features. This dimensional expansion provides more robust anomaly detection that is reliable for quality control while organizing the complexity in a structured manner through multi-scale or multi-feature analysis
Data Source
AI summary
Devices, systems, and methods obtain one or more training images; obtain a test image; select one or more associated pixels in the training images for a target pixel in the training images; calculate respective value relationships between a value of the target pixel and respective values of the associated pixels in the training images; select one or more associated pixels in the test image for a target pixel in the test image; and detect an anomaly in the target pixel in the test image based on the respective value relationships between the value of the target pixel and the respective values of the associated pixels in the training images and on respective value relationships between a value of the target pixel and respective values of the associated pixels in the test image.