Attention-Layer Image Classification Using Pixel Correlation Maps

Resolve Bottlenecks,
Find Innovative Solutions
Generate Solutions

Solution Overview

Problem

Existing image classifiers, such as those used in autonomous vehicles, often misclassify images, leading to unsafe device behavior due to insufficient performance.

Innovation Solution

Implementing an image classifier with an attention layer that utilizes a correlation matrix based on pixel dependencies, normalized using a predefined scheme, to enhance classification accuracy by focusing on relevant image regions.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Reliability

If conventional image classifiers are used, then device control is simple, but classification accuracy is low leading to unsafe behavior

Engineering Contradiction:
Improveclassification accuracyVSAvoidclassifier structure
Core Design Contradiction:
ReliabilityVSDevice complexity

Solution Approach 1:

The patent segments the image processing task by introducing an attention layer that divides the image into multiple regions and processes them separately using region-specific correlation matrices. This allows different parts of the image to be analyzed with appropriate computational resources, improving accuracy without uniformly increasing complexity across the entire system.

Inventive Principle:
Principle #1Segmentation

Solution Approach 2:

The patent adds an attention dimension to the traditional classification architecture by introducing attention weights and correlation matrices that operate in a new computational dimension. This allows the model to selectively focus on relevant regions without fundamentally restructuring the base classifier, thus improving reliability while controlling complexity growth.

Inventive Principle:
Principle #17Another dimension (Dimensionality change)

2Measurement precision

If attention layer with correlation matrix is added, then classification precision is improved, but computational resources and energy consumption increase

Engineering Contradiction:
Improveclassification precisionVSAvoidenergy consumption
Core Design Contradiction:
Measurement precisionVSUse of energy by moving object

Solution Approach 1:

The patent applies local quality by computing correlation matrices specifically for regions identified as important by the attention mechanism, rather than computing a single global correlation matrix for the entire image. This localized approach concentrates computational resources on critical regions, improving precision while reducing overall energy consumption compared to global processing.

Inventive Principle:
Principle #3Local quality

Solution Approach 2:

The attention layer implements partial action by selectively processing only certain regions of the image with full correlation matrix computation, while other regions receive simplified or no processing. This partial application of the computationally intensive operation maintains precision for critical classifications while reducing total energy consumption.

Inventive Principle:
Principle #16Partial or excessive action

Data Source

PatentEP3866071B1Device and method for classifying images using an attention layer
Publication Date: 2025.12.03 ROBERT BOSCH GMBH
  • EP3866071B1 patent drawingFigure 1
  • EP3866071B1 patent drawingFigure 2
  • EP3866071B1 patent drawingFigure 3

AI summary

A computer-implemented method for classifying images (x) using an image classifier (60) that comprises at least one attention layer (63), wherein the method comprises the following steps: • Providing an input activation map (X63) to the attention layer (63), wherein the input activation map (X63) is determined depending the image (x); • Providing, in the attention layer (63), a correlation matrix (C) comprising a plurality of entries (Ci,j), wherein each of the entries characterizes a relative dependence of a first pixel (ix, iy) of the input activation map (X63) on a second pixel (jx,jy) of the input activation map (X63), further wherein the correlation matrix (C) depends on a prior matrix (G); • Obtaining an attention layer output (Oh) depending on the input activation map (X63) and the correlation matrix (C); • Computing a classification (y) of the image (x) depending on the attention layer output (Oh).