Adaptive Reference Pixel Generation for Video Coding
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Conventional intra-prediction methods in video coding do not adequately consider the variation of reference pixel characteristics with coding conditions, leading to suboptimal prediction performance, particularly for diagonal prediction modes where interpolation filters with fixed tap lengths and coefficients fail to adapt to changing coding conditions.
Innovation Solution
An adaptive reference pixel generation process is introduced, where the tap length of the interpolation filter is dynamically set based on the block size and quantization parameter, allowing for the selection of shorter tap lengths for larger blocks and smaller quantization steps, and longer tap lengths for smaller blocks and larger quantization steps, to better match the characteristics of the decoded picture.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Device complexity
If fixed tap length and filter coefficients are used for interpolation, then device complexity is reduced, but prediction precision deteriorates due to inability to adapt to varying coding conditions
Solution Approach 1:
The patent applies dynamics by making the interpolation filter adaptable to different coding conditions. Specifically, the filter selectively uses different reference pixel generation methods based on the prediction mode and block size: for horizontal/vertical predictions it copies adjacent pixel values directly, while for diagonal predictions it performs interpolation using decoded pixel values. This dynamic adaptation resolves the contradiction by optimizing prediction precision for each condition without uniformly increasing device complexity.
Solution Approach 2:
The patent changes parameters based on coding conditions by adjusting the interpolation approach according to prediction mode and block size. For diagonal prediction modes, it changes from simple copying to interpolation using decoded pixel values, effectively changing the filter parameters (tap length and coefficients) to match the local characteristics of the decoded picture, thereby improving prediction precision without maintaining fixed parameters.
2Speed
If simple pixel copying is used for reference pixel generation, then processing speed is improved, but prediction precision deteriorates due to rough granularity in prediction direction
Solution Approach 1:
The patent dynamically selects between fast copying and more complex interpolation based on prediction mode. For horizontal and vertical predictions, it uses simple copying for speed. For diagonal predictions where granularity effects are more problematic, it switches to interpolation using decoded pixel values, achieving a dynamic balance between processing speed and prediction precision.
Solution Approach 2:
The patent applies local quality by using different reference pixel generation strategies for different prediction modes. Diagonal prediction modes, which are more susceptible to granularity effects, receive the more sophisticated interpolation treatment, while horizontal and vertical modes use the simpler copying method. This localized approach optimizes prediction precision where needed without unnecessarily complicating all prediction operations.
3Measurement precision
If adaptive reference pixel generation based on block size and quantization parameter is implemented, then prediction precision is improved, but device complexity increases
Solution Approach 1:
The patent implements dynamics by making the reference pixel generation process adaptive to block size and quantization parameter. The system dynamically determines whether to use simple copying or interpolation based on these parameters, optimizing prediction precision for different coding scenarios while avoiding unnecessary complexity for cases where simple methods suffice.
Solution Approach 2:
The patent changes the interpolation parameters (whether to interpolate and which reference pixels to use) based on block size and quantization parameter values. This parameter adaptation improves prediction precision by matching the reference pixel generation method to the local characteristics of the block being predicted, rather than using a fixed complex method for all cases.
Data Source
Figure 1A~1C
Figure 2
Figure 3
AI summary
An object is to reduce an intra-prediction error and improve coding efficiency by introducing an adaptive reference pixel generating process in accordance with coding conditions into intra-prediction. In picture encoding or picture decoding for generating a prediction signal using spatial inter-pixel prediction and encoding or decoding a picture using a prediction residual signal which is a difference between the prediction signal and an original signal, a tap length of an interpolation filter necessary for generating a reference pixel of intra-prediction is set based on one or both of a size of a block which is a processing unit of coding, transform, or prediction and a quantization parameter of the block for the reference pixel, a filtering process which generates the reference pixel is performed using the interpolation filter corresponding to the set tap length, and an intra-prediction signal corresponding to a designated intra-prediction mode is generated using the generated reference pixel.