Sequential Noise Filtering Using Mahalanobis Distances
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing image processing pipelines face challenges in efficiently reducing noise in image data, particularly impulse noise, as linear filters tend to blur images, while nonlinear filters are not effective in all cases, leading to resource-intensive CPU usage and increased power consumption.
Innovation Solution
The implementation of multiple noise models using Mahalanobis distances between pixels and their neighboring pixels, where coarse noise models are initially applied followed by finer noise models to iteratively filter the image data, enhancing noise reduction efficiency.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Productivity
If linear filters are used for noise reduction, then processing speed is maintained, but impulse noise cannot be effectively eliminated and image edges become blurred
Solution Approach 1:
The patent segments the noise reduction process into multiple stages: a first filtering pass using linear filters for general noise reduction, followed by a second filtering pass using nonlinear filters specifically targeted at impulse noise. This segmentation allows each filter type to operate optimally in its suitable context, maintaining processing speed while improving overall noise reduction effectiveness.
Solution Approach 2:
The patent applies different filtering characteristics to different regions and noise types within the image. Linear filters are applied first for general areas, then nonlinear filters are applied selectively to regions containing impulse noise. This local differentiation allows the system to maintain edge sharpness in most areas while effectively removing impulse noise where present.
2Reliability
If nonlinear filters are used for impulse noise reduction, then noise elimination effectiveness improves, but processing complexity and resource consumption increase
Solution Approach 1:
The patent divides the image processing into two sequential passes: first applying linear filters which are computationally simpler and faster, then applying nonlinear filters only in the second pass. This segmentation reduces overall processing complexity compared to applying nonlinear filters throughout, while still achieving effective impulse noise elimination where needed.
Solution Approach 2:
The patent applies nonlinear filtering partially rather than universally - the first filter pass handles general noise with simpler operations, and the second pass applies more complex nonlinear operations only where additional filtering is beneficial. This partial application reduces overall computational complexity while maintaining effective impulse noise removal.
3Reliability
If multiple filtering passes are applied sequentially, then noise reduction effectiveness improves, but processing time and resource consumption increase
Solution Approach 1:
The patent segments the multi-pass filtering process into distinct stages with different computational requirements. The first pass uses computationally efficient linear filters that can be applied quickly to the entire image, while the second pass uses more intensive nonlinear filters only where needed. This segmentation allows the system to achieve effective noise reduction while minimizing overall processing time.
Solution Approach 2:
The patent changes filtering parameters between passes - using linear filter parameters in the first pass and nonlinear filter parameters in the second pass. This parameter variation allows optimization of processing time in the first pass while maintaining noise reduction effectiveness in the second pass, balancing overall processing time with effectiveness.
Data Source
AI summary
Embodiments of the present disclosure relate to performing noise reduction on an input image by first filtering the input image based on coarse noise models of pixels and then subsequently filtering the filtered input image based on finer noise models. The finer noise models use the same or more number of neighboring pixels than the coarse noise filters. The first filtering and subsequent filtering of a pixel in the input image use Mahalanobis distances between the pixel and its neighboring pixels. By performing iterations of filtering using more refined noise models, the noise reduction in the input image can be performed more efficiently and effectively.