Image Gradient Distribution for Abnormality Detection Under Lighting Variation

Resolve Bottlenecks,
Find Innovative SolutionsGenerate Solutions

Solution Overview

Problem

Existing AI-based image analysis techniques for detecting abnormalities in objects require large amounts of learning data and are inaccurate in environments with fluctuating luminance values or colors, making them costly and unreliable.

Innovation Solution

An abnormality detection device that analyzes the gradient distribution of luminance in input images to determine the presence or absence of abnormalities without relying on AI, using gradient distribution generation and abnormality determination units to process images captured in varying lighting conditions.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Reliability

If AI-based image analysis is used to detect abnormalities, then detection capability is improved, but the requirement for large amounts of learning data increases, making the system costly and complex

Engineering Contradiction:

Improvedetection capabilityVSAvoidrequirement for learning data

Core Design Contradiction:

ReliabilityVSDevice complexity

Solution Approach 1:

The patent extracts and uses only the essential gradient distribution information from images, eliminating the need for complex AI models and large training datasets. By focusing solely on luminance gradient distributions in divided regions, the system achieves abnormality detection without requiring AI-based learning frameworks.

Inventive Principle:

Principle #2Taking out (Extraction)

Solution Approach 2:

The patent replaces expensive and complex AI models with a simple, lightweight gradient distribution analysis method. The system uses basic image processing techniques (dividing images into regions and calculating luminance gradients) that can be implemented with minimal computational resources and no extensive training data.

Inventive Principle:

Principle #27Cheap short-living objects (Disposable)

2Reliability

If AI-based image analysis is used to detect abnormalities, then detection capability is improved, but image capture and labeling by expert workers become costly

Engineering Contradiction:

Improvedetection capabilityVSAvoidcost of image capture and labeling

Core Design Contradiction:

ReliabilityVSEase of manufacture

Solution Approach 1:

The patent extracts only the necessary gradient distribution features from images, eliminating the need for expert workers to perform image capture and labeling. The system automatically analyzes luminance gradients in divided regions, replacing manual expert operations with automated computational analysis.

Inventive Principle:

Principle #2Taking out (Extraction)

3Ease of operation

If luminance value and color are used as basis for abnormality determination, then detection can be performed, but detection accuracy is limited in environments where luminance and color vary greatly

Engineering Contradiction:

Improveability to perform detectionVSAvoiddetection accuracy

Core Design Contradiction:

Ease of operationVSMeasurement precision

Solution Approach 1:

The patent changes the detection parameter from absolute luminance values and colors to relative luminance gradient distributions. By analyzing the distribution of gradient directions in divided regions rather than absolute brightness or color values, the system becomes invariant to environmental variations in lighting and color conditions.

Inventive Principle:

Principle #35Parameter changes

4Measurement precision

If gradient distribution analysis is used to detect abnormalities, then detection accuracy in fluctuating environments is improved, but the method becomes more complex

Engineering Contradiction:

Improvedetection accuracyVSAvoidcomplexity of gradient distribution analysis

Core Design Contradiction:

Measurement precisionVSDevice complexity

Solution Approach 1:

The patent segments the image into multiple predetermined regions and analyzes gradient distributions within each region. This segmentation approach simplifies the overall analysis by breaking down the complex task of image-wide gradient analysis into manageable regional analyses, making the method more implementable while maintaining high detection accuracy.

Inventive Principle:

Principle #1Segmentation

Data Source

PatentUS12561783B2Abnormality detection device, abnormality detection method, and abnormality detection system

Publication Date: 2026.02.24 KOKUSAI DENKI ELECTRIC INC

AI summary

A technique is provided for detecting the presence or absence of an abnormality with respect to an object appearing in a target image with high accuracy without using AI, even in an environment in which luminance values and colors are likely to fluctuate. An abnormality detection device (230) includes: an image input unit (232) for inputting an input image indicating an abnormality detection target object; a gradient distribution generation unit (234) for dividing the input image into predetermined regions and generating, for each region, a gradient distribution that indicates a distribution of a luminance gradient direction of the region; and an abnormality determination unit (236) for determining the presence or absence of an abnormality by analyzing the gradient distribution generated for each region.