Neural Network Scene Recognition Validation for Image Signal Processing

Resolve Bottlenecks,
Find Innovative Solutions
Generate Solutions

Solution Overview

Problem

Conventional scene recognition methods in mobile terminals, such as color channel-based or light measurement device-assisted methods, suffer from high false recognition rates, leading to low accuracy in image signal processing and affecting the quality of photos and videos.

Innovation Solution

The method employs a neural network to preliminarily recognize scenes and uses attribute information like light intensity and foreground location to determine the accuracy of the recognition, performing enhancement processing only when the scene is accurately identified, thereby improving recognition accuracy and image signal processing quality.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Reliability

If conventional scene recognition methods (color channel, template matching, or light measurement device) are used, then the device complexity is low, but the scene recognition accuracy deteriorates with high false recognition rate

Engineering Contradiction:
Improvescene recognition accuracyVSAvoiddevice complexity
Core Design Contradiction:
ReliabilityVSDevice complexity

Solution Approach 1:

The patent replaces conventional scene recognition methods (color channel analysis, template matching, light measurement device) with a neural network-based recognition system. This substitution transitions from traditional mechanical/optical measurement approaches to an intelligent algorithmic approach, significantly improving scene recognition accuracy while reducing false recognition rates. The neural network is trained to recognize various scene types and can accurately identify scenes even in challenging conditions where conventional methods fail.

Inventive Principle:
Principle #28Mechanics substitution (Replace mechanical system)

Solution Approach 2:

The patent introduces attribute information parameters (light intensity, foreground location) to validate and verify scene recognition results. By changing from single-parameter recognition to multi-parameter verification, the system improves reliability. The attribute information serves as additional validation criteria, allowing the system to confirm or correct neural network recognition results, thereby reducing false recognition while maintaining manageable device complexity.

Inventive Principle:
Principle #35Parameter changes

2Reliability

If scene recognition is performed using existing methods, then the processing speed is fast, but the image signal processing quality deteriorates due to low recognition accuracy

Engineering Contradiction:
Improveimage signal processing qualityVSAvoidprocessing time
Core Design Contradiction:
ReliabilityVSLoss of time

Solution Approach 1:

The patent performs preliminary scene recognition using the neural network before conducting detailed image signal processing. This preliminary action allows the system to quickly identify the scene type and apply appropriate processing parameters in advance, ensuring high-quality image signal processing from the outset. The attribute information verification is also performed preliminarily to confirm recognition accuracy before committing to specific processing paths, preventing time loss from reprocessing due to incorrect scene identification.

Inventive Principle:
Principle #10Preliminary action

Solution Approach 2:

The patent implements a feedback mechanism where attribute information (light intensity, foreground location) is used to verify and validate the neural network's scene recognition results. This feedback loop ensures that only accurately recognized scenes proceed to enhancement processing, while uncertain cases are re-evaluated or corrected. This feedback system maintains high processing quality by preventing incorrect scene-based processing while minimizing time loss through efficient verification rather than complete reprocessing.

Inventive Principle:
Principle #23Feedback

3Reliability

If enhancement processing is performed based on recognized scene, then the image signal processing quality is improved, but the device complexity increases due to additional processing steps

Engineering Contradiction:
Improveimage signal processing qualityVSAvoidprocessing complexity
Core Design Contradiction:
ReliabilityVSDevice complexity

Solution Approach 1:

The patent applies enhancement processing selectively based on the recognized scene type and validated attribute information. Different enhancement algorithms and parameters are applied to different scene types (e.g., portrait, landscape, night scene) rather than applying uniform processing to all images. This local quality approach ensures optimal processing quality for each specific scene while avoiding unnecessary processing complexity for scenes that don't require enhancement, thereby improving the quality-complexity ratio.

Inventive Principle:
Principle #3Local quality

Solution Approach 2:

The patent performs enhancement processing only when scene recognition accuracy is confirmed through attribute information validation. Rather than applying enhancement to all images, the system selectively applies processing only when necessary and appropriate, based on verified scene characteristics. This partial action approach improves overall processing quality by ensuring enhancement is applied only when it will be beneficial, while avoiding the complexity and potential degradation from unnecessary processing on already-adequate images.

Inventive Principle:
Principle #16Partial or excessive action

Data Source

PatentEP3674967B1Image signal processing method, apparatus and device
Publication Date: 2023.01.18 HUAWEI TECH CO LTD
  • EP3674967B1 patent drawingFigure 1
  • EP3674967B1 patent drawingFigure 2
  • EP3674967B1 patent drawingFigure 3

AI summary

This application provides an image signal processing method, apparatus, and device. The image signal processing method includes: obtaining an image signal, where the image signal is derived based on a sensor signal collected by an image sensor; preliminarily recognizing, by using a neural network, a scene to which the image signal belongs; and then further determining, by using attribute information of the image signal, whether the preliminarily recognized scene is accurate; and if determining that the scene is accurate, performing enhancement processing on the image signal based on the scene, to generate an enhanced image signal. This can increase scene recognition accuracy, and can further improve quality of image signal processing.