Image Segmentation Using Noise-Dependent Weights

Resolve Bottlenecks,
Find Innovative Solutions
Generate Solutions

Solution Overview

Problem

Existing image processing methods struggle with noise signals in medical and other fields, which are often considered nuisances and ignored, rather than being utilized to guide segmentation processes effectively, especially in images with pronounced spatial noise variations.

Innovation Solution

A method that incorporates spatially varying noise-dependent weights to control model adaptation during image segmentation, using noise levels to guide the optimization process and avoid highly noisy edges, by modifying the objective function to penalize locations with higher noise levels, thereby improving segmentation accuracy.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Measurement precision

If traditional segmentation methods are used that ignore noise signals, then the segmentation process is simpler, but segmentation accuracy deteriorates in images with spatially varying noise levels

Engineering Contradiction:
Improvesegmentation accuracyVSAvoidsegmentation process complexity
Core Design Contradiction:
Measurement precisionVSDevice complexity

Solution Approach 1:

The patent applies local quality by introducing spatially varying noise-dependent weights that adapt the segmentation model to local noise characteristics at different image locations. The objective function incorporates position-dependent weight maps that reflect local noise levels, allowing the segmentation to be more accurate in low-noise regions while being more conservative in high-noise regions, thereby resolving the contradiction between accuracy and complexity.

Inventive Principle:
Principle #3Local quality

Solution Approach 2:

The patent changes parameters by modifying the objective function to include noise-dependent weight parameters that vary spatially across the image. These weight parameters are derived from noise maps or variance maps and dynamically adjust the influence of image gradients in the segmentation process, improving accuracy without requiring a complete redesign of the segmentation framework.

Inventive Principle:
Principle #35Parameter changes

2Measurement precision

If noise signals are utilized to guide model adaptation, then segmentation accuracy improves in noisy regions, but the computational complexity increases

Engineering Contradiction:
Improvesegmentation accuracyVSAvoidcomputational power
Core Design Contradiction:
Measurement precisionVSPower

Solution Approach 1:

The patent applies preliminary action by pre-computing noise maps or variance maps before the segmentation process. These pre-computed noise characteristics are then used to generate weight maps that guide the model adaptation. By performing the noisy computation upfront and using the results to guide subsequent segmentation, the method improves accuracy while managing computational power requirements through staged processing.

Inventive Principle:
Principle #10Preliminary action

3Reliability

If the model adaptation is controlled by local noise levels, then the segmentation avoids highly noisy edges, but the model adaptation process becomes more complex

Engineering Contradiction:
Improvesegmentation reliabilityVSAvoidmodel adaptation complexity
Core Design Contradiction:
ReliabilityVSDevice complexity

Solution Approach 1:

The patent introduces an intermediary element - the noise-dependent weight map - that mediates between the image data and the model adaptation process. These weights act as a controlling mechanism that modulates the influence of image gradients based on local noise levels, allowing the model to adapt reliably to true edges while being protected from noise-induced false adaptations, without requiring complex decision logic.

Inventive Principle:
Principle #24Intermediary (Mediator)

Data Source

PatentEP3014573B1Methods of utilizing image noise information
Publication Date: 2017.12.13 KONINKLIJKE PHILIPS NV
  • EP3014573B1 patent drawingFigure 1
  • EP3014573B1 patent drawingFigure 2
  • EP3014573B1 patent drawingFigure 3

AI summary

Image processing methods and related apparatuses (SEG,UV). The apparatuses (SEG,UV) operate to utilize noise signal information in images (IM). According to one aspect, apparatus (SEG) uses the noise information (FX) to control a model based segmentation. According to a further aspect, apparatus (UV) operates, based on the noise information (FX), to visualize the uncertainty of image information that resides at edge portions of the or an image (IM).