Temporal Noise Reduction Using Future Frames
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Traditional noise reduction methods in video systems often fail to effectively reduce noise in images due to motion afterimages in low luminance environments, limiting the degree of noise reduction that can be achieved without introducing artifacts.
Innovation Solution
The method involves performing temporal noise reduction operations using not only past frames but also future frames, with the future frame being used in conjunction with infinite impulse response filtering and finite impulse response filtering to enhance noise reduction performance without causing motion afterimages.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If traditional temporal noise reduction operations use only past frames, then processing complexity is reduced, but noise reduction performance is limited and motion afterimages occur in low luminance environments
Solution Approach 1:
The patent performs preliminary noise reduction operations using past frames before the current frame is fully processed, preparing intermediate results that can be combined with future frames. This allows the system to leverage temporal correlations from multiple directions without waiting for all frames to be completely processed, thereby improving noise reduction performance while managing processing complexity through staged computation.
Solution Approach 2:
The patent transitions from unidirectional temporal processing (only past frames) to bidirectional temporal processing (past and future frames) by introducing a temporal dimension extension. This allows noise reduction operations to exploit correlations in both forward and backward time directions, significantly improving noise reduction performance while the structured algorithm manages the increased computational complexity.
2Reliability
If noise reduction degree is increased using traditional methods, then more noise is removed, but motion afterimages are introduced in low luminance environments
Solution Approach 1:
The patent introduces future frames as intermediary information to mediate between the current frame and past frames during noise reduction. By incorporating future frame data, the algorithm can distinguish between actual motion and noise more effectively, allowing higher noise reduction degrees to be achieved without introducing motion afterimages in low luminance environments.
Solution Approach 2:
The patent dynamically adjusts noise reduction parameters based on luminance conditions and temporal correlations from multiple frames. By changing the weighting parameters and filtering strength adaptively, the system can achieve higher noise reduction degrees in appropriate conditions while preventing motion afterimages through parameter modulation that responds to the actual image content and temporal patterns.
3Reliability
If future frames are used for noise reduction, then noise reduction performance is improved, but processing time and frame delay increase
Solution Approach 1:
The patent performs preliminary processing and preparation of future frames in advance, conducting initial noise reduction operations and generating intermediate results before they are needed for the final composite output. This preliminary action reduces the actual processing time when the frames must be combined, thereby improving noise reduction performance while minimizing the impact on frame delay and processing time.
Data Source
AI summary
In a method of processing an image, a first composite frame image corresponding to an n-th input frame image is generated by performing a first temporal noise reduction operation based on the n-th input frame image and an (n−m)-th input frame image. The (n−m)-th input frame image is received prior to the n-th input frame image. A second composite frame image corresponding to the n-th input frame image is generated by performing a second temporal noise reduction operation based on the n-th input frame image and an (n+k)-th input frame image. The (n+k)-th input frame image is received subsequent to the n-th input frame image. The second composite frame image is provided as an n-th output frame image corresponding to the n-th input frame image.