Machine Vision Anomaly Detection With Adaptive Cropping
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Conventional image analysis techniques for anomaly detection in machine vision applications require significant computational resources, fail to maintain contextual information, and struggle with novel defects, especially in multi-object images, and are prone to false positives due to low confidence scores, which waste resources and time in limited inspection scenarios.
Innovation Solution
A system and method utilizing image subtraction, adaptive cropping, and pseudo-one-class classification to identify and classify anomalies, employing aggregate thresholding and adaptive region cropping to enhance accuracy and reduce false positives.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If conventional image analysis techniques are used to thoroughly analyze each part of an input image, then anomaly detection accuracy is improved, but computational resources and time consumption increase significantly
Solution Approach 1:
The patent divides the image analysis process into multiple stages: initial full-image analysis to identify regions of interest, followed by focused detailed analysis only on those identified regions. This segmentation approach maintains high anomaly detection accuracy while significantly reducing overall computational resource consumption by avoiding exhaustive analysis of entire images.
Solution Approach 2:
The patent applies partial action by performing comprehensive analysis only on specific regions of interest rather than the entire image. The system identifies suspicious regions through initial screening and applies detailed analysis selectively, thereby achieving sufficient detection accuracy with reduced computational effort.
2Difficulty of detecting and measuring
If conventional image cropping techniques are used to extract anomalies, then localization is achieved, but contextual information such as size and dimension ratio is lost
Solution Approach 1:
The patent applies local quality by preserving different types of information at different scales: full-resolution contextual information is maintained in the original image coordinates, while localized anomaly details are extracted in cropped regions. This allows the system to have both precise localization and preserved contextual information like size and dimension ratios.
3Adaptability or versatility
If conventional classifiers are used on multi-object images, then object detection may occur, but anomaly detection fails due to lack of contextual understanding
Solution Approach 1:
The patent segments the image into multiple regions of interest and processes each region independently through the anomaly detection pipeline. This allows the system to handle multi-object images effectively by applying contextual analysis to each object or region separately, maintaining high anomaly detection accuracy even when multiple objects are present.
4Productivity
If low confidence score detections are treated as novel anomalies, then comprehensive detection is achieved, but false positives increase and waste computational resources
Solution Approach 1:
The patent implements feedback mechanisms where detection results are evaluated against multiple criteria including confidence scores, regional context, and anomaly patterns. Low confidence detections are not automatically classified as novel anomalies but are subjected to further verification, reducing false positives while maintaining comprehensive detection coverage.
Data Source
AI summary
Systems and methods for visual inspection and anomaly detection are provided herein. An inspection image is compared to a golden sample image to identify an anomaly map. Aggregate thresholding is performed on the anomaly map to identify anomalies. Adaptive cropping is performed on the identified anomalies to obtain cropped images of the anomalies. The cropped images are provided to an image classification model which is a pseudo one-class classifier. The image classification model classifies the anomalies.