Image data pre-processing for neural networks

Resolve Bottlenecks,
Find Innovative Solutions
Generate Solutions

Solution Overview

Problem

Neural networks face computational inefficiencies due to the need to process image data with varying noise levels and loss of exposure information, requiring them to perform noise level prediction and exposure determination, which increases network complexity and introduces errors.

Innovation Solution

An image signal processor (ISP) that separates and processes pixel data and characteristic data in parallel, applying corresponding operations to maintain exposure and noise information, reducing the computational burden on neural networks by providing optimized input data.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Measurement precision

If the neural network processes raw image data with varying noise levels and exposure information, then the neural network can perform comprehensive image analysis, but the computational complexity and processing time increase significantly

Engineering Contradiction:
Improveimage analysis accuracyVSAvoidneural network complexity
Core Design Contradiction:
Measurement precisionVSDevice complexity

Solution Approach 1:

The patent segments the image processing task into two distinct parts: (1) an image signal processor that performs noise reduction and exposure correction on raw image data, and (2) a neural network that receives the pre-processed data for analysis. This segmentation allows each component to specialize, reducing the overall complexity of the neural network while maintaining analysis accuracy.

Inventive Principle:
Principle #1Segmentation

Solution Approach 2:

The image signal processor performs preliminary actions (noise reduction and exposure correction) on the raw image data before it is fed into the neural network. By pre-processing the data to correct exposure variations and reduce noise, the neural network receives cleaner, standardized input, which reduces its computational complexity and processing requirements.

Inventive Principle:
Principle #10Preliminary action

2Adaptability or versatility

If the neural network processes image data with noise and exposure variations, then it can handle real-world image conditions, but the processing time and computational resources increase

Engineering Contradiction:
Improveimage condition handlingVSAvoidprocessing speed
Core Design Contradiction:
Adaptability or versatilityVSProductivity

Solution Approach 1:

The image signal processor performs preliminary noise reduction and exposure correction on raw image data before it is fed into the neural network. By pre-processing the data to correct exposure variations and reduce noise, the neural network receives cleaner, standardized input, which reduces its computational complexity and processing requirements, thereby improving processing speed.

Inventive Principle:
Principle #10Preliminary action

Solution Approach 2:

The image signal processor automatically adjusts noise levels and exposure parameters in the image data before processing, making the data suitable for neural network analysis without requiring the neural network itself to perform these adjustments. This self-service approach handles real-world image conditions efficiently.

Inventive Principle:
Principle #25Self-service

3Reliability

If the neural network performs noise level prediction and exposure determination, then it can compensate for image quality variations, but the network size and computational burden increase

Engineering Contradiction:
Improveimage processing reliabilityVSAvoidcomputational energy
Core Design Contradiction:
ReliabilityVSUse of energy by moving object

Solution Approach 1:

The patent segments the processing tasks by having the image signal processor handle noise reduction and exposure correction separately from the neural network's analysis functions. This segmentation eliminates the need for the neural network to perform noise level prediction and exposure determination, reducing its size and computational energy requirements while maintaining processing reliability through the dedicated pre-processing stage.

Inventive Principle:
Principle #1Segmentation

Data Source

PatentEP3761263B1Image data pre-processing for neural networks
Publication Date: 2026.01.28 ARM LTD
  • EP3761263B1 patent drawingFigure 1a~1b
  • EP3761263B1 patent drawingFigure 2
  • EP3761263B1 patent drawingFigure 3~4

AI summary

An image processor (100) comprising an input module (120) for receiving image data (110) from an image sensor; an image processing module (130) arranged to perform one or more operations on at least a portion of the image data (110) to generate processed image data (160i); and a characteristic processing module (140) arranged to perform one or more characteristic processing operations on at least a portion of the characteristic data to generate processed characteristic data (160n). The portion of the characteristic data is associated with the portion of the image data (110); the one or more characteristic processing operations are associated with the one or more image processing operations. The image processor (100) further comprises an output module (150) for outputting the processed image data (160i) and processed characteristic data (160n).