Image Decoding with Adaptive Thresholds for Edge-Aware Linear Prediction
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
The existing MMLM prediction method in image encoding and decoding systems has low edge-component sensitivity, leading to inadequate utilization of prediction coefficients for linear prediction.
Innovation Solution
An image decoding and encoding system that sets a threshold value based on the distribution or change in reference samples, allowing for the use of different prediction coefficients depending on whether the sample of the first component is equal to or above the threshold, thereby enhancing the suitability of prediction coefficients for linear prediction.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Ease of operation
If the average value of reference samples is used as the threshold value in MMLM prediction, then the calculation is simple, but the edge-component sensitivity is low and the prediction coefficient cannot be suitably used
Solution Approach 1:
The patent changes the threshold value parameter from a fixed average value to a dynamically determined value based on gradient calculations. By computing gradients between reference samples and comparing them to a gradient threshold, the system adapts the threshold to local image characteristics, thereby improving edge-component sensitivity while maintaining computational feasibility through gradient-based comparisons.
2Productivity
If a fixed prediction coefficient is used for all luminance sample values, then the processing is efficient, but the prediction accuracy for different regions (especially edge regions) is insufficient
Solution Approach 1:
The patent applies local quality by selecting different prediction coefficients based on local image characteristics. When the gradient between luminance samples exceeds a threshold, indicating an edge region, a first prediction coefficient is used; otherwise, a second prediction coefficient is applied. This allows the system to adapt to local variations in image content, improving prediction accuracy for both edge and non-edge regions while maintaining processing efficiency through gradient-based region classification.
Data Source
Figure 1
Figure 2
Figure 3
AI summary
An image decoding device includes: a decoding-side first generation unit that generates, for a target unit, a prediction sample of a first component; a decoding-side linear prediction unit that uses the sample of the first component and a prediction coefficient to generate a linear prediction sample of a second component; and a decoding-side second generation unit that uses the linear prediction sample of the second component to generate, for the target unit, a prediction sample of the second component. The decoding-side linear prediction unit uses a first coefficient as the prediction coefficient when the sample of the first component is equal to or below a threshold value, and uses a second coefficient which is different from the first coefficient as the prediction coefficient when the sample of the first component is greater than the threshold value. The threshold value is set on the basis of a parameter representing a distribution of or a change in a reference sample of the first component and/or the second component contained in a reference unit which is referenced for the target unit.