Image Normalization for Accurate Anomaly Detection Under Scene Changes
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing image recognition systems struggle to differentiate between normal changes due to environmental factors and actual abnormalities in monitored targets, making it difficult to accurately detect anomalies.
Innovation Solution
An information processing apparatus utilizing a learned model constructed through machine learning, specifically a Generative Adversarial Network (GAN), normalizes input images to match reference images, enabling detection of abnormalities by comparing normalized images to reference images using pixel-level differences and feature matching.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If image recognition is performed on captured images to detect abnormalities, then abnormality detection capability is improved, but false detection increases due to normal environmental changes
Solution Approach 1:
A learned model constructed from multiple normal images serves as an intermediary between the input image and the determination process. This model generates a normalized image that represents what a normal state should look like, allowing the system to compare the normalized image with the actual input image to detect abnormalities while filtering out normal environmental variations
Solution Approach 2:
The system changes the parameter representation of images by transforming the input image into a normalized form through the learned model. This transformation adjusts image parameters (such as lighting, color, and texture) to match the characteristics of normal images, enabling more accurate abnormality detection by focusing on genuine deviations rather than normal variations
2Measurement precision
If normalization process using learned model is applied, then detection accuracy is improved, but processing complexity increases
Solution Approach 1:
The learned model is constructed in advance through machine learning using multiple normal images. This preliminary action creates a pre-trained normalization model that can be directly applied to new images without requiring complex real-time analysis, thus improving detection accuracy while maintaining manageable processing complexity
Solution Approach 2:
The learned model creates a copy or representation of normal image characteristics. By generating a normalized image that copies the essential features of normal states, the system can compare this copy with the input image to detect abnormalities, simplifying the processing by working with standardized representations rather than raw diverse images
Data Source
AI summary
An appearance of a desired change can be detected in a more preferred manner even under circumstances where various kinds of changes can appear in a target image. An information processing apparatus includes: a generator configured to generate a normalized image by performing a normalization process using a learned model constructed in advance based on a machine learning with a plurality of normal images as input, on an input image; a determiner configured to determine whether the normalized image and a prescribed reference image approximately match or not; and an outputter configured to output notification information according to a result of the determination to a prescribed output destination.