Intra Planar Coding Prediction Using Pivot Reference Pixels
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current video coding systems face challenges in predicting pixels on non-reconstructed reference lines, particularly the right and bottom reference lines, which are not encoded and thus not available for decoding, leading to inefficiencies in interpolation and prediction processes.
Innovation Solution
The proposed solution involves a decoder predicting pixels on non-reconstructed reference lines by interpolating between corresponding pixels on reconstructed left and top reference lines, using pivot reference pixels and intercept reference pixels, and combining intra planar and interpolation prediction approaches to accurately estimate pixel values.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If interpolation is used to predict pixels on non-reconstructed reference lines, then prediction capability is improved, but computational complexity increases
Solution Approach 1:
The current block is divided into multiple prediction units (PUs), and each PU is further divided into sub-PUs for independent prediction. This segmentation allows the decoder to process smaller regions with different prediction modes, reducing the overall computational complexity while maintaining prediction accuracy for each segment.
Solution Approach 2:
The patent pre-calculates and stores gradient values (dx and dy) for different prediction modes in lookup tables during encoding. The decoder retrieves these pre-computed gradient values directly without performing complex calculations, significantly reducing computational complexity while maintaining prediction precision.
Solution Approach 3:
The patent applies full interpolation only when necessary (e.g., for non-reconstructed reference lines), while using simplified prediction methods for reconstructed lines. This selective application of complex processing reduces overall computational burden while maintaining accuracy where most needed.
2Measurement precision
If more reference lines are reconstructed and encoded, then prediction accuracy is improved, but transmission bandwidth increases
Solution Approach 1:
Instead of encoding all reference lines, the patent inverts the approach by only encoding necessary reference lines (left and top) and deriving others (right and bottom) through prediction. This reduces the amount of data transmitted while maintaining prediction capability through mathematical relationships between reference lines.
Solution Approach 2:
The patent introduces gradient calculations (dx and dy) as intermediary values that mediate between encoded reference lines and predicted pixels. These gradient intermediaries enable accurate prediction of non-reconstructed reference lines without transmitting additional reference line data, thus saving bandwidth while maintaining precision.
3Measurement precision
If multiple prediction approaches are combined, then prediction accuracy is improved, but processing complexity increases
Solution Approach 1:
The patent dynamically selects prediction modes and approaches based on the specific characteristics of each prediction unit and available reference data. The decoder adapts the level of processing complexity according to the actual needs of each block, using simple copying when reference lines are available and more complex interpolation only when necessary, thus balancing accuracy and processing load.
Data Source
AI summary
Intra planar approach(es) may be used to predict a pixel(s) in a current block. The current block may be associated with a reconstructed left reference line, a reconstructed top reference line, and an non-reconstructed reference line to be predicted. The reconstructed reference lines may have been decoded and may be available. The non-reconstructed reference lines to be predicted may include an non-reconstructed right and/or an non-reconstructed bottom reference lines. A pivot reference pixel may be identified and may be located on an extension of the reconstructed left and/or top reference lines. A reference pixel may be determined and may be located on the reconstructed top and/or left reference lines. Pixels on the non-reconstructed reference line(s) may be predicted based on the pivot reference pixel and the reference pixel. Pixels of the current block may be predicted using the predicted pixels on the right and the bottom reference lines.