Intra Prediction Error Correction Using Distance-Based Laplace Approximation
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing video compression methods face challenges in maintaining prediction performance as the distance between pixels increases, leading to decreased compression efficiency, particularly in intra prediction, where block size adjustments compromise between prediction accuracy and control information overhead.
Innovation Solution
A moving image coding device that calculates and reflects prediction errors based on the distance from a reference pixel using a prediction error statistic evaluation unit, feature amount calculation, and prediction value generation, employing a Laplace distribution to approximate and correct prediction values along intra prediction directions.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If the block size of a processing block is reduced to maintain prediction accuracy, then prediction performance is improved, but the number of blocks increases resulting in increased control information overhead
Solution Approach 1:
The patent applies local quality by making prediction weights variable rather than uniform. Different pixels within the same block are assigned different weights based on their distance from the reference pixel and the predicted distribution characteristics. This allows the system to maintain accurate prediction for distant pixels without reducing block size, as each pixel receives customized weight treatment according to its local position and statistical properties.
Solution Approach 2:
The patent changes the parameter of prediction weight from a fixed value to a variable parameter that depends on pixel position and predicted distribution. By modeling the predicted distribution of pixel values and using this to determine weights, the system adapts the prediction parameters dynamically based on the specific characteristics of each pixel location, thereby maintaining prediction accuracy without increasing control overhead.
2Ease of manufacture
If simple copy of reference value is used in intra prediction, then processing is simplified, but prediction performance decreases as distance from reference pixel increases
Solution Approach 1:
The patent changes the prediction parameter from a simple copy operation to a weighted combination where weights are determined by the predicted distribution of pixel values. This allows the system to maintain processing simplicity while improving prediction performance, as the weight calculation is based on statistical modeling rather than complex adaptive algorithms.
Solution Approach 2:
The patent substitutes the mechanical/simple copy operation with a statistical modeling approach. Instead of directly copying reference values, the system uses probability distribution modeling to determine optimal prediction weights, replacing direct mechanical copying with a more sophisticated statistical substitution that improves accuracy.
3Productivity
If distance-based weighting of reference pixel is applied, then compression performance improves for distant pixels, but the method uses inter coding frame information which cannot be applied to intra prediction
Solution Approach 1:
The patent segments the prediction process into independent components: predicting the distribution of pixel values within the current block and determining weights based on this local distribution. This segmentation allows the method to be applied to intra prediction by focusing solely on the current block's statistics rather than relying on inter-frame information, thereby maintaining compression performance while ensuring intra-prediction applicability.
Data Source
AI summary
A moving image coding device includes: a evaluation unit configured to approximate a parameter of a statistic in accordance with a distance from a reference pixel located along the intra prediction direction to a pixel to be processed; a feature amount calculation unit configured to calculate, for a pixel near the pixel to be processed, a feature amount of a prediction value that is based on the intra prediction direction; and a prediction value calculation unit configured to estimate a prediction error corresponding to the distance using the parameter of the statistic approximated by the evaluation unit, reflect the prediction error in the prediction value that is based on the intra prediction direction using the feature amount calculated by the prediction value feature amount calculation unit, and generate a new prediction value.