Noise Level Estimation Using Pseudo-Standard Deviation
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current noise assessment methods in image processing are subjective, time-consuming, and lack uniform criteria, and existing automatic techniques are not viable for iterative processing across various imaging modalities, often relying on complex noise models or being incompatible with noise reduction filters.
Innovation Solution
A method and system that use pseudo-standard deviation (PSD) and histogram analysis to determine noise levels automatically, focusing on featureless regions and soft tissue pixels, allowing for reliable noise assessment without human intervention and applicable to multiple imaging modalities.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If subjective visual evaluation is used for noise assessment, then operator judgment can be applied, but it is time-consuming, expensive, and lacks uniform criteria
Solution Approach 1:
The patent replaces the mechanical/manual process of subjective visual evaluation with an automated computational system. The noise assessment is performed by a processor that calculates a noise index from the image data, eliminating the need for operator intervention and significantly reducing evaluation time while maintaining consistent criteria through algorithmic processing
Solution Approach 2:
The system enables self-assessment of noise levels by automatically processing image data and generating noise index values without requiring external operator input. The computational method performs measurements autonomously, allowing the system to evaluate its own image outputs for noise characteristics
2Device complexity
If a single indiscriminate index is used for spatially variant noise, then calculation is simple, but the noise index is not fully representative of the entire image
Solution Approach 1:
The patent divides the image into multiple regions of interest (ROIs) and calculates separate noise index values for each region. This segmentation allows the system to capture spatially variant noise characteristics across different areas of the image, providing a more comprehensive representation while maintaining manageable calculation complexity through systematic processing of divided regions
Solution Approach 2:
The patent extends noise assessment from a single global value to multiple regional values across the image space. By introducing the spatial dimension of multiple ROIs, the system captures the variability of noise across different image regions, transforming the assessment from one-dimensional (single value) to multi-dimensional (spatially distributed values)
3Extent of automation
If twin image subtraction method is used for noise assessment, then automatic processing is achieved, but it requires two reconstructions from odd and even views which makes it not viable for iterative processing
Solution Approach 1:
The patent performs noise assessment on the original reconstructed image before any iterative noise reduction processing is applied. By calculating the noise index from the initial image data, the system establishes a baseline noise level that can be used to evaluate the effectiveness of subsequent iterative processing steps, enabling automation without requiring multiple reconstructions
4Measurement precision
If Laplacian method is used for noise assessment, then edge pixels are removed according to edge detection, but noise is measured from the manipulated image rather than the original image
Solution Approach 1:
The patent extracts noise information directly from the original image data without applying manipulative transformations like Laplacian filtering. By calculating the noise index from the raw reconstructed image, the system obtains an accurate representation of the actual noise present in the image while avoiding the complexity of additional processing steps and the risk of measuring artifacts rather than true noise
Data Source
AI summary
Noise assessment is important to image quality evaluation as well as image processing. For example, the noise level estimation is used as criteria for terminating an iterative noise reduction process. To determine a meaningful noise level, the pixels in featureless regions are separated from the rest of the image. A new concept of pseudo-standard deviation (PSD) is introduced to automatically determine simple and reliable noise level estimates. Furthermore, a histogram of PSD is constructed with fine bins to calculate the moving average of the histogram. The first peak in filtered histogram gives the most representative noise measure as a desired approximation of true standard deviation.