Image Encoding With Adaptive Intra Prediction And Multi-Type Transforms
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing video compression technologies struggle to efficiently encode and decode high-resolution and stereoscopic image content, particularly in terms of intra prediction and transform methods, leading to suboptimal encoding efficiency and accuracy.
Innovation Solution
The proposed method involves generating a prediction block based on predefined modes, determining an intra reference line, and performing intra prediction using a candidate group that includes a default mode and multiple MPM candidates, with optional offset adjustments, and applying multi-type transforms like DCT-2 and DST-7.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If conventional intra prediction methods are used, then encoding simplicity is maintained, but intra prediction accuracy deteriorates for high-resolution and stereoscopic images
Solution Approach 1:
The current block is divided into multiple sub-blocks, and different intra prediction modes are applied to different sub-blocks. This segmentation allows the patent to improve prediction accuracy for complex high-resolution and stereoscopic images by adapting different prediction strategies to different regions, while maintaining manageable complexity through systematic processing of each sub-block
Solution Approach 2:
The patent dynamically selects intra prediction modes and reference lines based on block characteristics, surrounding block modes, and gradient calculations. This dynamic adaptation enables the system to optimize prediction accuracy for each specific block rather than applying a static method uniformly, resolving the contradiction between accuracy and complexity
2Productivity
If a single transform type is used, then processing complexity is reduced, but transform efficiency deteriorates for different block types and sizes
Solution Approach 1:
The patent applies different transform types (DST-7, DCT-2, DCT-8) to different blocks based on their specific characteristics such as intra prediction mode, block size, and gradient properties. This local optimization ensures that each block undergoes the most appropriate transform for its characteristics, improving overall transform efficiency without requiring complex global decision-making
Solution Approach 2:
The patent changes transform parameters (transform type selection) based on block characteristics including prediction mode, block size, and gradient calculations. This parameter adaptation allows the system to optimize transform efficiency for different block types and sizes while maintaining a relatively simple decision framework based on predefined criteria
3Measurement precision
If conventional reference pixel configuration is used, then encoding simplicity is maintained, but intra prediction accuracy deteriorates for blocks with varying gradients
Solution Approach 1:
The patent performs preliminary calculations of gradients and determines optimal reference lines before executing intra prediction. By pre-calculating gradient information and identifying the most suitable reference lines in advance, the system improves prediction accuracy for blocks with varying gradients while keeping the overall process organized and manageable
Solution Approach 2:
The patent introduces gradient calculations and reference line selection as intermediary steps between the original block and the final prediction. These intermediaries provide additional information about block characteristics and guide the selection of optimal prediction parameters, improving accuracy without directly complicating the core prediction mechanism
Data Source
AI summary
Image encoding/decoding methods and apparatuses according to the present invention may generate a prediction block of a current block based on a pre-defined prediction mode, generate a transform block of the current block through predetermined transformation, and reconstruct the current block based on the prediction block and the transform block.