MIP Intra Prediction Mode Derivation for Efficient Image Coding
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
The existing image encoding/decoding technologies face inefficiencies when handling high-resolution and high-quality images, particularly in the use of Matrix-based Intra Prediction (MIP) mode, as it cannot be integrated into processes beyond intra prediction without modification.
Innovation Solution
The method generates a derived intra prediction mode through decoder-side intra mode derivation (DIMD) using a histogram of gradients, applied to matrix-based intra prediction (MIP) blocks, to determine a suitable transform set and intra prediction mode candidate list, enhancing encoding/decoding efficiency.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If MIP mode is used for intra prediction, then prediction accuracy is improved, but the mode cannot be integrated into processes beyond intra prediction without modification
Solution Approach 1:
The patent creates a derived intra prediction mode by copying the gradient calculation concept from direction-based modes and adapting it for MIP. The histogram of gradients is calculated from the MIP prediction block, and the most probable mode is derived from this histogram, effectively copying the adaptation mechanism while preserving MIP's prediction accuracy.
Solution Approach 2:
The patent changes the parameter representation by converting MIP's matrix operation results into a derived mode index based on gradient histogram analysis. This parameter transformation allows the MIP mode to be represented in a form that can be integrated into existing mode determination processes that expect direction-based mode indices.
2Measurement precision
If high-resolution and high-quality images are processed, then image quality is improved, but transmission and storage costs increase
Solution Approach 1:
The patent implements feedback by calculating the histogram of gradients from the MIP prediction block and using this feedback to determine the most probable intra prediction mode. This feedback mechanism allows the encoder to adaptively select the best mode based on actual image content characteristics, improving compression efficiency for high-resolution images.
Solution Approach 2:
The patent performs preliminary action by generating the derived intra prediction mode and determining the transform set before the actual encoding process. The gradient histogram is calculated in advance, and the most probable mode is determined beforehand, which streamlines the subsequent encoding steps and reduces overall processing costs for high-quality images.
3Adaptability or versatility
If direction-based intra prediction mode is used, then compatibility with existing processes is maintained, but prediction accuracy for complex patterns is limited
Solution Approach 1:
The patent introduces an intermediary mechanism by using the histogram of gradients as a bridge between MIP's matrix operations and the traditional direction-based mode selection process. The gradient histogram serves as an intermediary representation that captures the directional information needed for mode selection while being derived from MIP's more accurate prediction block.
Data Source
AI summary
An image encoding/decoding method and apparatus, a recording medium storing a bitstream and a transmission method are provided. The image decoding method may comprise generating a derived intra prediction mode by performing decoder side intra mode derivation (DIMD) on a current block and storing the derived intra prediction mode. The current block is in a matrix based intra prediction (MIP) mode.