Image Encoding Intra Prediction With DIMD-Planar Blending
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing image compression technologies face challenges in accurately predicting pixel values within images, particularly in high-resolution and high-quality formats like HD and UHD, which affects the efficiency and accuracy of image encoding and decoding processes.
Innovation Solution
An intra prediction method and device based on Directional Mode Derivation (DIMD) that modifies prediction samples using a weighted sum between first and second prediction samples, determined by gradients in neighboring block areas, and adaptively selects planar modes based on gradient amplitudes and directionality.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If traditional intra prediction methods are used, then the encoding process is simple, but the prediction accuracy for high-resolution images is insufficient
Solution Approach 1:
The patent changes the parameters of prediction by deriving multiple intra prediction modes (including planar modes and directional modes) based on gradient calculations from neighboring blocks. Instead of using a single fixed prediction mode, the system dynamically selects and combines multiple modes with different parameters to improve prediction accuracy for high-resolution images.
Solution Approach 2:
The patent creates a composite prediction approach by combining multiple prediction samples (first prediction sample from DIMD mode and second prediction sample from planar mode) through weighted summation. This composite method leverages the strengths of different prediction modes to achieve superior prediction accuracy compared to individual modes alone.
2Measurement precision
If multiple prediction modes are derived and combined, then the prediction accuracy improves, but the computational complexity increases
Solution Approach 1:
The patent applies partial action by selectively deriving only the necessary prediction modes based on gradient conditions. Instead of computing all possible prediction modes for every block, the system calculates gradients and derives modes only when they meet certain conditions, reducing unnecessary computational power consumption while maintaining prediction accuracy.
Solution Approach 2:
The patent applies local quality by determining different prediction modes for different regions (sub-areas) of the current block based on local gradient characteristics. The neighboring area is divided into sub-areas, and each sub-area can have its own gradient calculation and mode selection, allowing the system to adapt to local image features without uniformly processing the entire block.
3Adaptability or versatility
If gradient-based mode derivation is used, then the adaptability to image content improves, but the processing time increases
Solution Approach 1:
The patent applies preliminary action by pre-calculating gradients from neighboring blocks before deriving prediction modes. The gradient information is computed in advance and stored, then used to determine the appropriate prediction modes. This preliminary gradient calculation avoids repeated computations during the prediction mode selection process, reducing overall processing time.
Solution Approach 2:
The patent segments the neighboring area into multiple sub-areas (left, top, top-left) and calculates gradients for each sub-area separately. This segmentation allows the system to efficiently process different regions with potentially different gradient characteristics, improving content adaptability while managing processing time through localized calculations.
Data Source
Figure 1
Figure 2
Figure 3
AI summary
An image encoding/decoding method and device, according to the present disclosure, can derive an intra prediction mode of the current block, derive a first prediction sample of the current block on the basis of the intra prediction mode, derive a second prediction sample of the current block on the basis of a predetermined planar mode, and derive a prediction sample of the current block on the basis of a weighted sum between the first prediction sample and the second prediction sample.