Multi-pass video noise filtering reduces ghosting artifacts
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Temporal noise filtering in video sequences often results in ghosting artifacts, especially at high filter strengths, due to the persistence of object contours in moving areas, which existing motion adaptive filters fail to completely eliminate.
Innovation Solution
A multi-pass application of a spatial noise filter on temporal differences is implemented, enhancing the robustness against ghosting artifacts by effectively using a temporal noise filter with a large kernel size through a small kernel filter, and computing blending factors based on motion values to generate a filtered frame.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If a temporal IIR noise filter is applied with high filter strength, then noise filtering effectiveness is improved, but ghosting artifacts increase
Solution Approach 1:
The patent segments the temporal filtering process into multiple passes, where each pass processes temporal differences at different levels. The first pass processes original temporal differences, while subsequent passes process differences of differences, effectively dividing the filtering operation into manageable stages that reduce ghosting artifacts while maintaining noise reduction effectiveness.
Solution Approach 2:
The patent introduces an additional temporal dimension by processing temporal differences at multiple levels. Instead of a single-pass filter, the system creates a multi-pass filter that operates on temporal differences, then on differences of differences, adding a temporal layer to the filtering process that eliminates ghosting artifacts.
2Object-generated harmful factors
If a motion adaptive temporal IIR noise filter is used, then ghosting artifacts are reduced, but some artifacts persist and computational complexity increases
Solution Approach 1:
The patent extracts the motion adaptation mechanism from the temporal filtering process and integrates it into the spatial filtering domain. By computing motion vectors and using them to guide spatial filter application, the system eliminates the need for complex temporal domain motion adaptation while achieving similar ghosting reduction effects.
Solution Approach 2:
The patent introduces motion vectors as an intermediary element that bridges spatial and temporal filtering. Motion vectors computed from frame differences serve as guidance for applying spatial filters, mediating between the need for temporal noise reduction and spatial detail preservation without requiring complex temporal motion adaptation.
3Reliability
If a spatial noise filter with large kernel size is applied, then noise filtering effectiveness is improved, but computational complexity increases
Solution Approach 1:
The patent segments the spatial filtering operation into multiple passes, where each pass applies a smaller kernel filter to different components of the image. By dividing the filtering process into sequential passes rather than using a single large kernel, the system achieves equivalent noise reduction with lower computational complexity at each step.
Data Source
AI summary
A method of noise filtering of a digital video sequence is provided that includes computing a motion image for a frame, wherein the motion image includes a motion value for each pixel in the frame, and wherein the motion values are computed as differences between pixel values in a luminance component of the frame and corresponding pixel values in a luminance component of a reference frame, applying a first spatial noise filter to the motion image to obtain a final motion image, computing a blending factor image for the frame, wherein the blending factor image includes a blending factor for each pixel in the frame, and wherein the blending factors are computed based on corresponding motion values in the final motion image, generating a filtered frame, wherein the blending factors are applied to corresponding pixel values in the reference frame and the frame, and outputting the filtered frame.