Information Processing for Size-Based Adversarial Patch Detection

Resolve Bottlenecks,
Find Innovative Solutions
Generate Solutions

Solution Overview

Problem

Existing methods for detecting adversarial patch attacks in image AI for object detection, such as at self-checkouts and license plate recognition, fail to accurately identify and mitigate the impact of adversarial patches due to random size and position of added missingness, leading to overlooked fraud and misrecognition.

Innovation Solution

An information processing program that determines a missing rate based on the minimum size of an adversarial patch, generates an image with missingness exceeding this rate, and compares detection results between original and modified images to improve detection accuracy.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Ease of manufacture

If missingness with random size and position is added to the input image, then the image processing can be performed, but the missingness may fail to cover the adversarial patch and fail to reduce the effectiveness of the patch

Engineering Contradiction:
Improveimage processingVSAvoidpatch attack detection
Core Design Contradiction:
Ease of manufactureVSReliability

Solution Approach 1:

The patent changes the parameters of missingness generation by determining a missing rate based on the minimum size of the adversarial patch and generating missingness that exceeds this rate. This ensures that the missingness is sufficiently large to cover the patch, transforming the random missingness into targeted missingness that reliably detects patch attacks.

Inventive Principle:
Principle #35Parameter changes

2Ease of operation

If random missingness is added to detect patch attacks, then detection method can be implemented, but detection accuracy is insufficient due to failure to cover the patch

Engineering Contradiction:
Improvedetection method implementationVSAvoidpatch attack detection accuracy
Core Design Contradiction:
Ease of operationVSMeasurement precision

Solution Approach 1:

The patent determines a missing rate based on the minimum size of the adversarial patch and generates missingness exceeding this rate. This parameter change ensures that the missingness covers the patch effectively, significantly improving detection accuracy while maintaining ease of implementation through automated parameter calculation.

Inventive Principle:
Principle #35Parameter changes

3Reliability

If missingness exceeding the missing rate is added to the first image, then the adversarial patch can be effectively covered, but the complexity of image processing increases

Engineering Contradiction:
Improveadversarial patch coverageVSAvoidimage processing complexity
Core Design Contradiction:
ReliabilityVSDevice complexity

Solution Approach 1:

The patent calculates the missing rate based on the minimum size of the adversarial patch and generates missingness that exceeds this rate. By using a clear mathematical criterion (missing rate = f(minimum patch size)), the patent achieves reliable patch coverage while keeping the processing logic simple and automated, avoiding unnecessary complexity.

Inventive Principle:
Principle #35Parameter changes

Data Source

PatentEP4589455A1Information processing program, information processing method, and information processing device
Publication Date: 2025.07.23 FUJITSU LTD
  • EP4589455A1 patent drawingFigure 1
  • EP4589455A1 patent drawingFigure 2
  • EP4589455A1 patent drawingFigure 3

AI summary

An information processing program causes a computer (10) to execute a process including acquiring a first value representing a size of a region of an object included in a first image, determining a missing rate for an adversarial patch, based on a second value representing a minimum size of the adversarial patch acquired according to the first value, generating a second image in which missingness exceeding the missing rate is added to the first image, and comparing a first detection result obtained by inputting the first image into an object detection model with a second detection result obtained by inputting the second image into the object detection model.