Linear Intra Prediction for Higher-Accuracy Image Compression
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing image compression technologies face challenges in achieving high prediction accuracy and efficient compression performance, particularly in high-resolution and high-quality image encoding/decoding processes.
Innovation Solution
The method and apparatus derive a linear model based on neighboring reconstructed samples and configure reference samples for intra prediction, considering the shape of the block, to enhance prediction accuracy and compression performance.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If conventional intra-prediction technology is used, then the prediction process is simple, but prediction accuracy is insufficient for high-resolution images
Solution Approach 1:
The patent changes the parameters of the prediction process by introducing linear model parameters (a, b, c) that can be derived from neighboring reconstructed samples. These parameters allow the prediction to adapt to different image characteristics and block shapes, improving accuracy without requiring completely new prediction mechanisms. The linear model parameters are calculated based on the variance in horizontal and vertical directions of neighboring samples.
Solution Approach 2:
The patent introduces an intermediary approach by using linear model parameters as a bridge between conventional intra-prediction and more complex prediction methods. The linear model acts as an intermediate that captures statistical relationships from neighboring samples and applies them to current block prediction, achieving improved accuracy while maintaining computational efficiency.
2Measurement precision
If more diverse prediction methods are used, then prediction accuracy increases, but the complexity of the encoding/decoding process increases
Solution Approach 1:
The patent introduces linear model parameters (a, b, c) that can be derived from neighboring reconstructed samples. These parameters allow the prediction to adapt to different image characteristics and block shapes, improving accuracy without requiring completely new prediction mechanisms. The linear model parameters are calculated based on the variance in horizontal and vertical directions of neighboring samples.
Solution Approach 2:
The patent makes the prediction process dynamic by allowing the linear model parameters to be adaptively derived from neighboring samples rather than using fixed prediction formulas. The parameters a, b, and c can vary depending on the image content and block characteristics, enabling the system to dynamically adjust to different prediction scenarios while maintaining a relatively simple computational structure.
3Measurement precision
If reference samples are configured without considering block shape, then the configuration process is simple, but prediction effectiveness decreases
Solution Approach 1:
The patent applies local quality by configuring reference samples differently based on the local block shape and characteristics. The linear model parameters are derived specifically from neighboring reconstructed samples that are relevant to the current block's geometry. For example, when the block is rectangular, the patent uses variance calculations in horizontal and vertical directions to select appropriate reference samples, thereby improving prediction effectiveness for each local configuration.
Solution Approach 2:
The patent changes the parameters of reference sample configuration by using linear model parameters (a, b, c) that are derived from neighboring reconstructed samples. These parameters allow the prediction to adapt to different image characteristics and block shapes, improving accuracy without requiring completely new prediction mechanisms. The linear model parameters are calculated based on the variance in horizontal and vertical directions of neighboring samples.
Data Source
AI summary
A method and apparatus for decoding/encoding an image according to the present disclosure may configure a reference sample for linear model intra prediction based on reconstructed neighboring samples of a current block, derive linear model parameters based on the reference sample, generate a prediction sample of the current block based on the linear model parameters, and decode/encode the current block based on the prediction sample of the current block.