Rolling Intra Prediction for Discontinuity-Resistant Video Coding
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing intra prediction schemes in image and video coding, such as those in HEVC, rely on linear displacement-based spatial prediction, which can introduce discontinuities and inefficiencies, particularly when encoding blocks with sharp edges, leading to difficulties in entropy coding non-zero coefficients at medium or high frequencies.
Innovation Solution
A modified intra prediction process, known as rolling intra prediction, where predicted pixels are used as inputs to generate other pixels within the same block, allowing for parallel processing of prediction and residual block reconstruction, independent of decoder reconstruction.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If linear displacement-based spatial prediction is used, then intra prediction can be performed using neighboring reconstructed pixels, but discontinuities are introduced and compression efficiency deteriorates
Solution Approach 1:
The patent segments the prediction process into two distinct phases: (1) generating initial predicted pixels using linear displacement-based spatial prediction from neighboring reconstructed pixels, and (2) generating additional predicted pixels using the already-generated predicted pixels within the same block. This segmentation allows each phase to leverage its strengths while mitigating weaknesses, resolving the contradiction between prediction reliability and compression efficiency.
Solution Approach 2:
The patent applies preliminary action by first generating a set of predicted pixels using neighboring reconstructed pixels before generating the remaining predicted pixels. This preliminary prediction creates a foundation that reduces discontinuities when subsequent pixels are generated, thereby improving both prediction reliability and compression efficiency without requiring full decoder reconstruction.
Data Source
Figure 1
Figure 2
Figure 3~4D
AI summary
Methods and devices for image and video coding using rolling intra prediction. The encoder selects a mode for a current block, the current block having a plurality of neighboring reference samples from an adjacent block, wherein the mode is associated with a prediction function. A predicted block of predicted pixels is constructed by initially determining at least some of the predicted pixels based on the prediction function with one or more of the neighboring reference samples as inputs to the prediction function, and by determining at least some other of the predicted pixels based on the prediction function with the at least some of the predicted pixels as inputs. The process proceeds to build the predicted block in a traversing order using predicted pixels of the same predicted block as inputs to the prediction function for determining other predicted pixels of the predicted block. The decoder performs the same prediction process.