Digital Image Anomaly Detection Using Semantic Class Similarity
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing methods for anomaly detection in digital images struggle to accurately distinguish between normal and abnormal object combinations, particularly in complex scenarios such as autonomous driving and manufacturing, due to a lack of effective utilization of semantic similarity between classes of objects.
Innovation Solution
A method and device that determine semantic similarity scores between classes of objects in a digital image using a knowledge graph, calculate a metric based on these scores, and detect anomalies or normalities by comparing the metric to a threshold, enhancing detection accuracy through weighted means or extremal scores.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If existing anomaly detection methods are used without semantic similarity information, then the detection process is simpler and faster, but the detection accuracy is insufficient particularly in complex scenarios
Solution Approach 1:
The patent pre-computes and stores semantic similarity scores between object classes in a knowledge graph before the actual anomaly detection process. This preliminary action allows the detection system to quickly retrieve pre-calculated semantic relationships during inference, improving detection accuracy without adding significant computational complexity during the actual detection phase
Solution Approach 2:
The patent introduces a knowledge graph as an intermediary structure that stores semantic similarity information between object classes. This knowledge graph acts as a mediator between object detection results and anomaly classification, providing pre-computed semantic relationships that enhance detection accuracy without requiring complex real-time computations
2Measurement precision
If semantic similarity scores are calculated for all object pairs, then the anomaly detection accuracy improves, but the computational complexity and processing time increase
Solution Approach 1:
The patent performs the computationally intensive task of calculating semantic similarity scores for all object class pairs in advance, before actual anomaly detection is needed. These pre-computed scores are stored in a knowledge graph structure, allowing rapid retrieval during detection without repeating the expensive calculations
Solution Approach 2:
The patent calculates semantic similarity scores for all possible object class pairs (excessive action) during the pre-computation phase, ensuring that comprehensive semantic information is available. During actual detection, only the relevant pre-computed scores need to be retrieved and applied, significantly reducing processing time
Data Source
AI summary
A device and a computer implemented method for processing a digital image for anomaly or normality detection. The method includes providing the digital image, determining, depending on the digital image, a first class for a first object and a second class for a second object depicted in the digital image, determining a score depending on semantic similarity between the first class and the second class, and detecting an anomaly or a normality depending on the score.