Accumulated Noise Mapping for Adaptive Image Noise Reduction
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing image processing systems suffer from inaccurate noise level estimation due to limited measurement points, leading to excessive noise filtration or insufficient noise reduction, which degrades image quality by losing fine details or introducing artifacts.
Innovation Solution
An accumulated noise model represented by a value map is used to calibrate noise reduction filtering based on actual noise levels across different stages of the image processing pipeline, adjusting filter strengths for each pixel or group of pixels.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If linear interpolation is used to estimate noise levels for pixels between measured points, then noise estimation can be performed across the entire image, but accuracy of noise estimates deteriorates leading to excessive noise filtration or insufficient noise reduction
Solution Approach 1:
The image is divided into multiple regions of interest (ROIs) with different characteristics (e.g., uniform regions, edge regions, texture regions). Each ROI is processed separately with dedicated measurement points and interpolation methods, allowing accurate noise estimation for each region without requiring full-image coverage measurement points.
Solution Approach 2:
Different noise estimation methods are applied to different regions of the image based on their specific characteristics. Uniform regions use one interpolation method while edge and texture regions use different methods, optimizing noise estimation accuracy for each local area rather than using a single global approach.
2Object-affected harmful factors
If excessive noise filtration is applied to processed image data, then noise artifacts are reduced, but fine details in the image are lost
Solution Approach 1:
The noise filtration strength is adjusted locally for each region of the image based on its characteristics. Edge regions and texture regions receive stronger noise filtration to remove artifacts, while uniform regions receive weaker filtration to preserve fine details, avoiding the one-size-fits-all approach.
Solution Approach 2:
The noise reduction process is made dynamic by adjusting filtration parameters in real-time based on the local characteristics of each image region. The system adapts the noise filtration strength dynamically for each ROI, applying stronger filtration where needed and weaker filtration where details must be preserved.
3Loss of information
If insufficient noise reduction is applied to processed image data, then fine details are preserved, but excessive noise artifacts degrade image quality
Solution Approach 1:
The system applies different noise reduction strengths to different image regions. Texture regions and edge regions receive sufficient noise reduction to eliminate artifacts, while uniform regions receive minimal noise reduction to preserve fine details, achieving optimal balance across the entire image.
Solution Approach 2:
The noise reduction process dynamically adjusts filtration parameters based on real-time analysis of each region's characteristics, ensuring that sufficient noise reduction is applied where artifacts are present while preserving details where noise is minimal.
Data Source
AI summary
This disclosure provides systems, methods, and devices for image signal processing that support improved noise reduction. In a first aspect, a method of image processing includes receiving an input image frame captured by an image sensor; receiving a value map corresponding to the input image frame; processing the input image frame to determine a processed image frame, wherein the processing comprises determining an updated value map based on the processing of the input image frame; and applying a noise reduction filter to the processed image frame based on the updated value map, wherein a strength of the noise reduction filter applied to each pixel of the input image frame is based on a corresponding value of the updated value map. Other aspects and features are also claimed and described.