Spatiotemporal Combining for Video Noise Reduction
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing video enhancement algorithms face high computational complexity and artifacts due to poor temporal registration, making noise reduction in video processing inefficient, especially under varying lighting conditions.
Innovation Solution
A low complexity, robust spatiotemporal combining method using block-based noise estimation, temporal weighting factors, Infinite Impulse Response (IIR) filters, and spatially adaptive weight selection to reduce noise while preserving the underlying signal.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Device complexity
If simple spatial averaging or wavelet domain methods are used for noise reduction, then computational complexity is reduced, but noise reduction effectiveness and signal preservation deteriorate
Solution Approach 1:
The video sequence is segmented into blocks, and noise estimation is performed block-by-block using local statistics. This allows the algorithm to process smaller regions independently, reducing overall computational complexity while maintaining effective noise reduction through localized adaptation to different noise characteristics in different regions.
Solution Approach 2:
The patent applies spatially adaptive filtering where the filtering strength and characteristics vary locally across different regions of the video frame based on local noise variance estimation. This enables effective noise reduction in each local region while preserving important local signal characteristics, achieving good noise reduction effectiveness without requiring globally complex processing.
2Reliability
If advanced de-noising algorithms with large spatiotemporal support are used, then noise reduction quality improves, but computational complexity increases significantly
Solution Approach 1:
The patent performs preliminary noise variance estimation for each block before applying the filtering operation. This preliminary action allows the algorithm to adapt the filtering parameters in advance, ensuring high noise reduction quality is achieved while avoiding the need for repeated complex computations during the actual filtering stage, thus reducing overall computational complexity.
Solution Approach 2:
The algorithm applies filtering selectively based on local noise characteristics rather than uniformly across the entire video frame. By performing partial action only where needed (in blocks with significant noise), the algorithm achieves effective noise reduction quality while reducing computational complexity compared to applying advanced algorithms uniformly across all regions.
3Reliability
If motion compensated temporal filtering is used, then temporal noise reduction improves, but artifacts due to poor temporal registration increase
Solution Approach 1:
The patent dynamically adjusts the temporal filtering parameters based on the estimated noise variance and local signal characteristics. By changing the filtering strength and temporal support parameters adaptively, the algorithm achieves effective temporal noise reduction while avoiding excessive filtering that would cause temporal artifacts, thus resolving the contradiction between noise reduction and artifact generation.
4Reliability
If intelligent weight determination is used to combine pixels, then noise reduction effectiveness improves, but computational complexity increases due to large spatiotemporal support volume
Solution Approach 1:
The patent divides the video into blocks and performs weight determination independently for each block based on local noise characteristics. This segmentation reduces the computational burden of weight determination by limiting the spatiotemporal support volume to local regions, while still achieving effective noise reduction through adaptive weighting within each block.
Solution Approach 2:
The algorithm adapts the weighting parameters based on locally estimated noise variance for each block. By changing the weighting parameters locally rather than using fixed or globally adaptive weights, the algorithm achieves effective noise reduction with reduced computational complexity, as each block's weights are determined independently using only local information.
Data Source
AI summary
A system and method for a low complexity and robust spatiotemporal combining for video enhancement is disclosed. In one embodiment, the method includes computing a standard deviation estimate between a video frame and a temporally neighboring frame of the video frame in a video sequence, computing an error value, for each sub-block of pixels within a block of pixels in a current video frame, between pixel values within the sub-block in the current video frame and corresponding motion compensated pixel values in a temporally neighboring video frame of the current video frame, computing a temporal weighting factor for each sub-block of pixels as a function of the error value and the standard deviation estimate, and combining the block of pixels in the current video frame and their corresponding motion compensated pixel values in the temporally neighboring video frame using the computed temporal weighting factor.