Regression Affine Candidates for Complex Motion Video Blocks
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing video coding technologies face challenges in improving coding efficiency, particularly in handling complex motion patterns beyond simple translations, such as zoom, rotation, and perspective motions, which are common in real-world scenarios.
Innovation Solution
The proposed method determines regression affine candidates based on motion fields of neighboring coding units or blocks, including adjacent and non-adjacent positions, temporal and spatial locations, and history tables, to enhance coding efficiency in video processing.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Productivity
If conventional video coding technologies are used, then the coding process is simple, but the coding efficiency is insufficient especially for complex motion patterns
Solution Approach 1:
The patent segments the motion field determination process into multiple stages by considering different types of coding units (CU) at different hierarchical levels. It divides motion prediction into spatial and temporal dimensions, using adjacent spatial CUs and temporal CUs separately, then combines them through regression affine models. This segmentation allows complex motion patterns to be handled systematically without overwhelming complexity in a single step.
Solution Approach 2:
The patent performs preliminary actions by pre-determining motion fields of multiple coding units before processing the current video block. It collects motion fields from adjacent spatial CUs, temporal CUs, and history tables in advance, then uses these pre-computed motion fields to derive regression affine candidates. This preliminary collection and organization of motion field data enables efficient processing of complex motion patterns without real-time computational burden.
2Measurement precision
If motion fields of multiple coding units are collected from various positions, then the regression affine candidate is more accurate, but the processing time increases
Solution Approach 1:
The patent applies local quality by selectively collecting motion fields from different spatial and temporal locations based on their relevance to the current video block. It prioritizes adjacent spatial CUs and temporal CUs that are most likely to provide accurate motion information, rather than uniformly processing all possible CUs. The regression affine model weights and selection criteria ensure that locally relevant motion fields contribute more significantly, improving precision while reducing unnecessary processing time.
Solution Approach 2:
The patent employs partial action by determining motion fields of a selected subset of coding units rather than all possible CUs. It collects motion fields from adjacent spatial positions, temporal positions, and history tables, but selectively processes only the most relevant ones based on availability and relevance criteria. This partial collection approach achieves sufficient precision for accurate motion prediction while significantly reducing processing time compared to exhaustive collection.
3Adaptability or versatility
If regression affine model is used to determine motion fields, then complex motion patterns are handled better, but the computational complexity increases
Solution Approach 1:
The patent implements universality by using a single regression affine model framework that handles multiple motion patterns (translation, zoom, rotation, perspective) through unified mathematical operations. The model takes motion fields from different CU types and positions as input and automatically adapts to represent various motion characteristics through the regression coefficients. This multi-functional approach eliminates the need for separate processing pipelines for different motion types, improving adaptability while controlling computational complexity through a single versatile algorithm.
Solution Approach 2:
The patent applies parameter changes by using the regression affine model to compute motion fields as functions of spatial and temporal parameters. Instead of directly calculating complex motion transformations, it expresses motion fields through regression coefficients that capture motion patterns. The model changes parameters (motion field values) based on input motion fields from different CUs, transforming them into accurate representations of complex motion. This parameter-based approach simplifies computation compared to direct geometric transformations while maintaining high adaptability to various motion patterns.
Data Source
AI summary
Embodiments of the present disclosure provide a solution for video processing. A method for video processing is proposed. In the method, for a conversion between a current video block of a video and a bitstream of the video, motion fields of a plurality of coding units coded before the current video block is determined. At least one of the plurality of coding units is collected from at least one of: an adjacent neighboring position, an adjacent neighboring position at a location, a collocated temporal position, an adjacent temporal position, a non-adjacent spatial position, a non-adjacent temporal position, or a history table of the current video block. A regression affine candidate of the current video block is determined based on the motion fields of the plurality of coding units. The conversion is performed based on the regression affine candidate.