Non-Straight Intra-Prediction for Video Block Encoding
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
High Efficiency Video Coding (HEVC) and its successor, JVET Exploratory Model (JEM), face limitations in predicting blocks with non-straight features in video coding, as the existing angular intra-prediction modes only extend pixels along straight lines, which can lead to suboptimal compression efficiency for blocks with curved or complex textures.
Innovation Solution
Introducing a new intra-prediction mode that extends neighboring pixels along non-straight lines, using pre-defined curves or piece-wise line constructions to better match the texture features in video blocks, thereby improving prediction accuracy and reducing the bits required for residual coding.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Productivity
If angular intra-prediction modes are used to extend pixels along straight lines, then the prediction is simple and computationally efficient, but the prediction accuracy deteriorates for blocks with curved or complex textures
Solution Approach 1:
The patent applies curvature by extending reference pixels along curved lines instead of straight lines. The curved extension lines better match the texture features in video blocks with curved patterns, improving prediction accuracy while maintaining computational efficiency through pre-defined curve configurations
Solution Approach 2:
The patent introduces dynamic adaptability by allowing the selection between straight-line and curved-line extension modes. The encoder can dynamically choose the appropriate extension type based on the local texture characteristics, optimizing prediction accuracy for different block types while maintaining simplicity for uniform regions
2Measurement precision
If the number of intra-prediction modes is increased from 35 to 66 angular modes, then the granularity of prediction directions is improved, but the complexity of mode selection and signaling increases
Solution Approach 1:
The patent segments the intra-prediction modes into two distinct categories: straight-line extension modes and curved-line extension modes. This segmentation allows the system to maintain the existing 35 straight-line modes for simple cases while adding a separate set of curved-line modes for complex textures, avoiding the need to renumber and reorganize all 66 modes into a single complex selection framework
Solution Approach 2:
The patent creates a universal intra-prediction framework that can handle both straight-line and curved-line extensions through a unified mode selection mechanism. The system maintains compatibility with existing HEVC decoders by using the same mode indexing for straight-line modes while adding curved-line modes as an extension, allowing the same decoder architecture to handle both types without requiring fundamental redesign
3Measurement precision
If pre-defined curves are used for pixel extension, then the prediction accuracy for curved features is improved, but the bitrate for signaling curve parameters increases
Solution Approach 1:
The patent applies preliminary action by pre-defining a set of standard curve configurations before encoding. These pre-defined curves are stored in both encoder and decoder, eliminating the need to signal curve parameters during encoding. The encoder simply selects from the pre-defined set, and the decoder reconstructs using the same pre-defined curves, thus avoiding bitrate overhead for parameter transmission
Solution Approach 2:
The patent uses copying by having the decoder copy the pre-defined curve configurations from the encoder's perspective. Both encoder and decoder maintain identical copies of the pre-defined curve set, ensuring that the same curve parameters are used for prediction without needing to transmit them in the bitstream, thus achieving lossless curve parameter sharing
Data Source
AI summary
A method of encoding a block of a picture into a video bitstream is provided. The method is performed by a video encoder and comprises encoding a first block of the picture, determining an intra-prediction mode for a second block of the picture, for which second block the intra-prediction is made from pixels in the first block along a non-straight line, encoding the second block using the determined intra-prediction mode, and signaling syntax elements for the second block in the video bitstream.