CIIP Image Encoding With Adaptive Weights and Block Partitioning
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing video coding technologies face challenges in efficiently processing increasing amounts of digital video data, particularly in improving prediction accuracy and encoding/decoding processes.
Innovation Solution
An image encoder and decoder that utilize intra and inter prediction methods, combining prediction blocks using weighted mixing of intra and inter prediction samples, and flexible partitioning of blocks into rectangular and non-rectangular shapes to enhance prediction accuracy.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If intra prediction and inter prediction are combined with weighted mixing, then prediction accuracy is improved, but encoding complexity increases
Solution Approach 1:
The patent implements dynamic selection of prediction modes (intra, inter, or CIIP) and adaptive weighting factors based on block characteristics and content analysis. The encoder dynamically determines whether to apply CIIP and adjusts weights w1 and w2 based on the specific block, allowing the system to adapt to different video content types and motion patterns, thereby improving prediction accuracy without always incurring maximum complexity
Solution Approach 2:
The patent applies different prediction strategies to different blocks within a picture based on local characteristics. Each block can independently use intra prediction, inter prediction, or CIIP with customized weights, allowing high-quality prediction where needed while maintaining lower complexity in simpler regions, thus resolving the contradiction between overall accuracy improvement and global complexity increase
2Measurement precision
If blocks are partitioned into rectangular and non-rectangular shapes, then prediction accuracy is improved, but processing complexity increases
Solution Approach 1:
The patent segments video blocks into different shapes (rectangular and non-rectangular including triangular partitions) to better match the actual content structure. By dividing blocks according to edge directions and motion patterns, the segmentation allows more accurate prediction regions while maintaining manageable processing through systematic partitioning rules
Solution Approach 2:
The patent introduces asymmetric non-rectangular partitions (such as triangular partitions with different orientations) to match asymmetric motion patterns and edge directions in video content. This asymmetric partitioning improves prediction accuracy for diagonal edges and complex motion while the systematic approach to generating these partitions keeps processing complexity controlled
3Measurement precision
If CIIP is performed with different weights for intra and inter prediction, then prediction accuracy is improved, but encoding time increases
Solution Approach 1:
The patent performs preliminary analysis of block characteristics (edge directions, motion patterns, content type) before determining the prediction mode and weights. This preliminary action allows the encoder to quickly identify candidates for CIIP and pre-determine appropriate weight ranges, reducing the time required for optimization while maintaining high prediction accuracy
Solution Approach 2:
The patent uses parameter changes in the form of adaptive weighting factors (w1, w2) that are adjusted based on block characteristics rather than exhaustively searching all possible weight combinations. This parameter adaptation approach maintains high prediction accuracy by optimizing weights for each block type while significantly reducing encoding time compared to exhaustive search methods
Data Source
AI summary
An image encoder includes: circuitry; and a memory coupled to the circuitry. The circuitry, in operation: calculates first values of a current block using intra prediction, the intra prediction being limited to planar mode, the planar mode using multiple reference pixels for each pixel location of the current block; calculates second values of the current block using inter prediction; calculates third values of the current block by weighting the first values and the second values; and encodes the current block using the third values, and in the calculating of the third values, a first weight is applied to the first values and a second weight is applied to the second values, the second weight being different from the first weight.