Adversarial Patch Generation Using Region Search for Flexible Attacks

Resolve Bottlenecks,
Find Innovative Solutions
Generate Solutions

Solution Overview

Problem

Existing patch-based adversarial attack methods for computer vision neural networks are limited by fixed positions and shapes of adversarial patches, leading to poor attack performance and inefficiency in evaluating the safety of these networks, particularly in physical scenarios.

Innovation Solution

Optimize the positions, shapes, and textures of adversarial patches by segmenting images into regions and using discrete search techniques to select target regions, allowing for flexible and effective adversarial patch generation.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Device complexity

If fixed positions and shapes of adversarial patches are used, then the device complexity is reduced, but the attack performance deteriorates

Engineering Contradiction:
Improvepatch configuration complexityVSAvoidattack performance
Core Design Contradiction:
Device complexityVSReliability

Solution Approach 1:

The patent makes the patch positions and shapes dynamic by optimizing them according to the image content and target object locations. Instead of using fixed predetermined positions, the system dynamically determines optimal patch configurations based on the specific image being attacked, thereby improving attack performance while maintaining manageable complexity through automated optimization

Inventive Principle:
Principle #15Dynamics

Solution Approach 2:

The patent changes the parameters of adversarial patches (positions, shapes, sizes) from fixed values to optimized variables. By treating patch parameters as adjustable variables that can be optimized through discrete search methods, the system achieves better attack performance without requiring complex manual configuration

Inventive Principle:
Principle #35Parameter changes

2Adaptability or versatility

If discrete search is used to select target regions, then the adaptability to different images is improved, but the computing time increases

Engineering Contradiction:
Improveimage adaptabilityVSAvoidcomputing time
Core Design Contradiction:
Adaptability or versatilityVSLoss of time

Solution Approach 1:

The patent segments the image into multiple candidate regions before performing discrete search. By dividing the image into predefined regions of interest, the search space is significantly reduced compared to searching all possible positions, thereby maintaining high adaptability while reducing computing time

Inventive Principle:
Principle #1Segmentation

Solution Approach 2:

The patent performs discrete search only on selected candidate regions rather than exhaustively searching the entire image space. This partial search approach focuses computational resources on the most promising areas, achieving good adaptability with reduced computing time

Inventive Principle:
Principle #16Partial or excessive action

Data Source

PatentUS12597246B2Method and apparatus for generating adversarial patch
Publication Date: 2026.04.07 ROBERT BOSCH GMBH
  • US12597246B2 patent drawing
  • US12597246B2 patent drawing
  • US12597246B2 patent drawing

AI summary

A method for generating a set of adversarial patches for an image. The method includes segmenting the image into a plurality of regions; selecting a set of target regions that satisfies an attacking criterion by discretely searching of the plurality of regions; and generating a set of adversarial patches by using the set of target regions.