Image Noise Reduction via Localized Pixel Filtering

Resolve Bottlenecks,
Find Innovative Solutions
Generate Solutions

Solution Overview

Problem

Existing noise reduction methods in images struggle to effectively reduce noise while preserving picture details and edges, particularly in scenarios where computational resources are limited, such as in real-time digital camera applications.

Innovation Solution

A method involving low-pass filtering and a linear combination of pixel values, where the modification of pixel values is based on thresholds and an attenuation function to differentiate between noise reduction and preservation of details, using a processor to compute distances and apply appropriate modifications to pixel values in a picture component.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Object-affected harmful factors

If aggressive noise reduction is applied to remove more noise, then noise visibility is reduced, but picture details and edges are overly smoothed

Engineering Contradiction:
Improvenoise visibilityVSAvoidpicture detail preservation
Core Design Contradiction:
Object-affected harmful factorsVSManufacturing precision

Solution Approach 1:

The patent applies different processing strengths to different regions of the image by detecting edge pixels versus non-edge pixels. Non-edge pixels undergo stronger noise reduction using linear combination with low-pass filtered values, while edge pixels are preserved with minimal modification. This local differentiation allows aggressive noise reduction in smooth areas without blurring important picture details and edges.

Inventive Principle:
Principle #3Local quality

2Object-affected harmful factors

If complex noise reduction algorithms are used to improve noise reduction effectiveness, then noise is reduced more effectively, but computational power and time requirements increase

Engineering Contradiction:
Improvenoise reduction effectivenessVSAvoidcomputational complexity
Core Design Contradiction:
Object-affected harmful factorsVSDevice complexity

Solution Approach 1:

The patent segments the image processing task into distinct steps: low-pass filtering to generate smoothed values, edge detection to identify pixels near boundaries, and conditional noise reduction application. This segmentation allows the use of simple, computationally efficient operations (low-pass filtering, threshold comparison) rather than complex algorithms, making the solution suitable for real-time processing on resource-constrained devices like digital cameras.

Inventive Principle:
Principle #1Segmentation

Solution Approach 2:

The patent applies noise reduction selectively rather than uniformly across all pixels. By computing a linear combination only for non-edge pixels and leaving edge pixels unchanged, the method performs partial action that achieves effective noise reduction in areas where it is needed (smooth regions) while avoiding unnecessary computational overhead in areas where it would be harmful (edge regions).

Inventive Principle:
Principle #16Partial or excessive action

Data Source

PatentEP3203439B1Method and device for reducing noise in a component of a picture
Publication Date: 2019.10.02 INTERDIGITAL VC HOLDINGS INC
  • EP3203439B1 patent drawingFigure 1
  • EP3203439B1 patent drawingFigure 2~3
  • EP3203439B1 patent drawingFigure 4

AI summary

The present principles relates to a method and a device for reducing noise in a component (C) of a picture, characterized in that the method comprises: - obtaining (100) a low-pass filtered component (Cf) by low-pass filtering said component (C) of the picture; - for a current pixel (i) in said component of the picture: - for at least one current neighboring pixel (j) of the current pixel (i), computing (110) a distance (D1), relative to the current neighboring pixel (j), between the value (Cf(i)) of the current pixel in the low-pass filtered component and the value (Cf(j)) of the current neighboring pixel (j) in the low-pass filtered component; - when the distances relative to said at least one neighboring pixels of the current pixel are lower than a first threshold (TH1) (120), modifying (130) the value (C(i)) of the current pixel in said component of the picture according to a linear combination of the value (C(i)) of the current pixel in said component of the picture and the value (Cf(i)) of the current pixel in the low-pass filtered component, - otherwise the value of the current pixel in said at least one component of the picture remains unchanged.