Bilateral Filter Weighting for Mosquito Artefact Reduction
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Mosquito and ringing artefacts, caused by strong pixel value variations such as edges in images, are difficult to reduce and can be annoying for viewers.
Innovation Solution
A method that generates weights based on pixel value differences and histogram densities, uses a bilateral filter to combine these weights, and mixes filtered pixel values with original pixel values to reduce artefacts.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Object-affected harmful factors
If a bilateral filter is applied to reduce mosquito and ringing artefacts, then the artefact reduction is improved, but the device complexity and computational requirements increase
Solution Approach 1:
The patent changes the parameters of the bilateral filter by using histogram-based weights instead of traditional spatial and intensity weights. The histogram is divided into multiple bins representing different intensity ranges, and weights are calculated based on the distribution of pixel intensities within these bins. This parameter transformation allows the filter to adaptively respond to local intensity variations caused by compression artifacts while maintaining computational feasibility through the binned histogram approach.
Solution Approach 2:
The patent applies local quality by calculating occurrence weights based on the local histogram distribution around each pixel. The soft-binned histogram approach creates overlapping intensity ranges that capture local pixel value distributions. This allows the filtering to be adapted to local image characteristics - in regions with compression artifacts, the histogram will show specific patterns that generate appropriate weights for artifact reduction, while preserving edges and important features where the histogram distribution differs.
2Object-affected harmful factors
If pixel values are averaged towards a group of similar-leveled pixels, then the artefact reduction is improved, but the detail preservation may be compromised
Solution Approach 1:
The patent transforms the filtering approach by using histogram binning with soft boundaries. Instead of hard thresholding or fixed-range averaging, pixels are assigned to overlapping intensity bins with gradual transitions. The occurrence weights are derived from the normalized histogram distribution, creating a smooth weighting function that adapts to local intensity characteristics. This parameter transformation enables the filter to distinguish between artifact-related intensity variations and genuine image details based on their statistical properties in the histogram domain.
Solution Approach 2:
The patent implements feedback through the histogram-based weight calculation mechanism. The histogram is computed from the input pixel values, and this histogram information feeds back into the weight calculation for the bilateral filter. The occurrence weights are derived from the histogram distribution, creating a closed-loop system where the filtering behavior adapts to the actual pixel value distribution in the local neighborhood. This feedback mechanism allows the filter to automatically adjust its strength based on local image characteristics, reducing artifacts in uniform regions while preserving edges and details where the histogram shows high-frequency content.
Data Source
AI summary
In one aspect, a method to reduce compression artefacts, includes generating a first set of weights based on pixel value differences, producing a soft binned histogram from values in a neighborhood of pixels, generating an adapted histogram from the soft binned histogram using interpolation, generating occurrence density weights from the adapted histogram, combining the first set of weights and the occurrence density weights, filtering pixel values a bilateral filter using the combined weights, mixing the filtered pixel values and original pixel values, and outputting the mix of filtered pixel values and original pixel values.