Video Denoising via Adaptive Spatial-Temporal Filtering
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing video denoising methods are prone to unreliable filtering results due to the use of local pixel data, especially in scenarios with object motion and occlusion, as they do not consider the reliability of pixel data and fail to accurately estimate signal statistics for filter determination.
Innovation Solution
A robust spatial-temporal hybrid filter that extracts current and previous pixel data sets to perform temporal filtering, estimates outlier quantities, and switches to spatial filtering if outliers exceed a threshold, using weighted averaging and M-estimator methods to downweight or exclude unreliable data, ensuring more reliable filtering results.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If traditional temporal filtering is used with motion compensation, then video denoising performance is improved, but reliability deteriorates when object motion and occlusion cause outliers in pixel data
Solution Approach 1:
The patent applies preliminary action by performing outlier detection and estimation before the filtering operation. The system estimates the quantity of outliers in the n-point pixel data set using M-estimator methods, and based on this preliminary assessment, decides whether to proceed with temporal filtering or switch to spatial filtering. This preliminary evaluation prevents unreliable data from degrading the filtering results.
Solution Approach 2:
The patent implements dynamics by making the filtering approach adaptive rather than fixed. The system dynamically switches between temporal filtering and spatial filtering based on the estimated outlier quantity. When outliers are detected (quantity >= threshold), the system transitions from temporal to spatial filtering, and vice versa. This dynamic adaptation ensures reliability across different video content conditions.
2Reliability
If spatial filtering is used to handle outliers, then filtering reliability is improved, but computational complexity increases due to additional processing steps
Solution Approach 1:
The patent applies partial action by performing outlier estimation only to the extent necessary to make a filtering mode decision. Rather than fully processing both spatial and temporal filtering paths, the system estimates outlier quantity and selectively applies only the appropriate filtering method. This partial processing approach maintains reliability while controlling computational complexity.
Solution Approach 2:
The patent introduces an intermediary component - the outlier estimation mechanism using M-estimators - that mediates between the raw pixel data and the filtering operation. This intermediary assesses data quality and directs the flow to appropriate filtering paths, adding a layer of complexity that enables reliable automatic adaptation without requiring manual intervention or overly complex rule-based systems.
3Manufacturing precision
If robust outlier estimation is implemented, then denoising quality is improved, but processing time increases due to additional computational steps
Solution Approach 1:
The patent performs preliminary outlier estimation using computationally efficient M-estimator methods before committing to a filtering approach. By quickly assessing the outlier quantity in the pixel data set and comparing it against a threshold, the system makes a rapid decision about which filtering path to follow, minimizing the time penalty while maintaining denoising quality.
Solution Approach 2:
The patent changes the parameter of filtering mode (spatial vs. temporal) based on the estimated outlier characteristics. Rather than always applying the most computationally intensive robust filtering method, the system adjusts the filtering parameter dynamically - using temporal filtering when data is clean and spatial filtering when outliers are present. This parameter adaptation maintains denoising quality while optimizing processing time.
Data Source
AI summary
A method for video denoising, the method includes extracting a current pixel from a current video frame and n−1 pixels from respective n−1 previous video frames to form an n-point pixel data set, wherein the current pixel has an initial value; temporally filtering the current pixel using the n-point pixel data set to generate a temporal estimated true value for the pixel; estimating a quantity of outliers in the n-point pixel data set; if the quantity of outliers is greater than or equal to a threshold, spatially filtering the current pixel using neighboring pixels in the current video frame to generate an spatial estimated true value for the pixel, and using the spatial estimated true value for the pixel as a final estimated value for the pixel; if the quantity of outliers is less than the threshold, using the temporal estimated true value for the pixel as the final estimated value for the pixel.