Asymmetric Deblocking Filter for Video Encoder Line Buffer Reduction
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current video encoding and decoding technologies face challenges in efficiently deblocking video sequences, particularly in handling asymmetric filter requirements and interactions with long filters and sub-block deblocking, especially when dealing with max CTU line buffers and sub-block motion compensation.
Innovation Solution
The proposed method introduces asymmetric deblocking filters that modify a limited number of samples on one side using 5 or 7 sample filters, depending on the block sizes and alignment, to reduce line buffers and enable parallel deblocking of CU boundaries, particularly when sub-block motion compensation is used.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Manufacturing precision
If asymmetric deblocking filters are applied to handle different block sizes and alignments, then deblocking precision is improved, but device complexity increases
Solution Approach 1:
The patent applies different filter lengths (5-sample or 7-sample filters) depending on the specific boundary conditions, block sizes, and alignments. Instead of using a uniform filter approach, the system adapts the filter characteristics to local requirements, applying stronger filtering where needed and weaker filtering where sufficient, thereby improving overall deblocking precision while managing complexity through selective application
Solution Approach 2:
The deblocking process is segmented into different cases based on block sizes and alignments. The patent divides the boundary processing into multiple scenarios (e.g., different CTU line buffer configurations, sub-block motion compensation cases) and applies appropriate filter lengths to each segment, allowing precise handling of each case without requiring a single complex filter design
2Reliability
If more line buffers are used to handle all boundary cases, then reliability is improved, but memory usage increases
Solution Approach 1:
The patent dynamically adjusts the number of line buffers required based on the specific boundary conditions and block configurations. By using asymmetric filter lengths adapted to each case, the system reduces the maximum line buffer requirement while maintaining reliability for all boundary types. The filter adapts its reach according to the actual needs of each boundary scenario rather than preparing for the worst case in all situations
Solution Approach 2:
The patent changes the filter length parameter (5-sample or 7-sample) based on the boundary characteristics, block sizes, and alignments. This parameter adaptation allows the system to maintain reliable deblocking for various boundary cases while reducing the memory resources required for line buffers, as shorter filters need fewer buffered samples to operate effectively
3Manufacturing precision
If sequential deblocking is used to ensure proper filtering order, then manufacturing precision is improved, but productivity decreases
Solution Approach 1:
The patent performs preliminary classification of boundary types and determines the appropriate filter length before actual filtering. By pre-assessing block sizes, alignments, and boundary characteristics, the system prepares the filtering strategy in advance, allowing for optimized parallel execution of multiple filtering operations without compromising the required filtering sequence and accuracy
Data Source
AI summary
An asymmetric deblocking method for deblocking a boundary between a P block and a Q block such that 5 samples within the P block and 7 or 3 samples within the Q block are modified. The method includes determining a value refP based on at least p5, determining a value refQ based on at least qx, wherein qx is q3 or q7; determining a value refMiddle based on at least p0 and q0, wherein p0 is directly adjacent to the boundary and q0 is directly adjacent to the boundary; performing a linear interpolation between refP and refMiddle; and performing a linear interpolation between refQ and refMiddle.