Multi-stage Coding Block Partition Search for Video Encoding
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current video encoding techniques face challenges in balancing compression performance with encoder complexity and computation time, particularly when using a large number of partition types, which can lead to increased processing time without significant compression gains.
Innovation Solution
A multi-stage coding block partition search method is employed to reduce the search space by eliminating some partition types and focusing on square partitions in the first stage, followed by a detailed search in the second stage, allowing for efficient selection of the optimal partition type between partition-none and partition-split types, and potentially third partition types, to encode video frames effectively.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Manufacturing precision
If a large number of partition types are used for video block prediction, then compression performance is improved, but encoder complexity and computation time increase significantly
Solution Approach 1:
The patent segments the partition type selection process into two distinct stages: a first stage that evaluates only partition-none and partition-split types, and a second stage that evaluates additional third partition types. This segmentation reduces the computational burden in each stage while maintaining comprehensive evaluation, thereby resolving the contradiction between compression performance and encoder complexity.
Solution Approach 2:
The patent performs preliminary evaluation of partition-none and partition-split types in the first stage before considering third partition types in the second stage. This preliminary action allows the encoder to quickly eliminate obviously suboptimal options and focus computational resources on more promising partition types, reducing overall complexity while preserving compression performance.
2Manufacturing precision
If a large number of partition types are used for video block prediction, then compression performance is improved, but computation time increases by 40%
Solution Approach 1:
The patent divides the computation into two time-separated stages: the first stage computes costs for partition-none and partition-split types, and the second stage computes costs for third partition types only if needed. This segmentation reduces computation time by 40% while maintaining compression performance by performing evaluations in a structured, incremental manner.
Solution Approach 2:
The patent performs preliminary cost comparisons between partition-none and partition-split types in the first stage before committing to evaluating third partition types in the second stage. This preliminary action eliminates unnecessary computations and reduces overall computation time while ensuring that compression performance is not compromised by the selective evaluation approach.
3Measurement precision
If all partition types are evaluated for every block, then optimal partition selection is achieved, but encoder complexity increases
Solution Approach 1:
The patent segments the partition type evaluation into two groups: common partition types (partition-none and partition-split) evaluated in the first stage, and additional third partition types evaluated in the second stage. This segmentation maintains partition selection accuracy by ensuring all types are eventually considered, while reducing encoder complexity through staged evaluation that avoids unnecessary computations.
Solution Approach 2:
The patent performs preliminary evaluation of the most common partition types before considering less frequent third partition types. This preliminary action maintains selection accuracy by ensuring that the majority of cases are handled efficiently in the first stage, while the second stage provides comprehensive evaluation only when necessary, thereby reducing overall encoder complexity.
Data Source
AI summary
Multi-stage coding block partition search is disclosed. A method includes selecting a partition-none partition type and a partition-split partition type for predicting the block, determining a first cost of predicting the block using the partition-none partition type, and determining a second cost of predicting the block using the partition-split partition type. The partition-none partition type and the partition-split partition type are selected from a set of partition types that includes the partition-none partition type, the partition-split partition type, and third partition types. The method also includes, on condition that the result meets a criterion, determining a respective encoding cost corresponding to at least some of the third partition types; selecting a selected partition type corresponding to a minimal cost amongst the partition-none partition type and the at least some of the third partition types; and encoding, in a compressed bitstream, the selected partition type.