Video Picture Prediction Padding for Out-of-Boundary Motion Blocks
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing video coding standards like HEVC and VVC face inefficiencies in motion-compensated temporal prediction when motion vectors point to reference blocks outside the boundaries of a reference picture, leading to suboptimal handling of out-of-boundary prediction blocks.
Innovation Solution
A method for video encoding and decoding that involves padding reference blocks based on motion vectors derived from nearest subblocks within the current block, particularly for uni-directional and bi-directional motion-compensated predictions, to improve the handling of reference blocks that extend beyond picture boundaries.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Area of stationary object
If motion vectors point to reference blocks outside the boundaries of a reference picture, then the prediction can cover a larger area, but the prediction quality deteriorates due to insufficient or invalid reference samples
Solution Approach 1:
The patent applies preliminary action by padding the reference picture with predicted sample values before the actual prediction process. The reference picture is extended with synthesized boundary samples derived from motion compensation of neighboring blocks, ensuring that when motion vectors point outside the original boundaries, valid reference samples are already in place. This pre-padding operation prevents prediction quality deterioration while maintaining the ability to cover larger areas.
Solution Approach 2:
The patent introduces an intermediary mechanism by using motion-compensated prediction from neighboring blocks to generate the padding samples. Instead of directly using invalid out-of-boundary samples or simple repetition, the invention creates intermediate predicted samples that serve as a bridge between the available reference picture content and the required extended prediction area. These intermediary samples are generated by applying motion compensation to neighboring blocks within the reference picture.
2Productivity
If padding is performed using simple repetition or zero-values, then the process is simple and fast, but the prediction quality is insufficient
Solution Approach 1:
The patent replaces the simple mechanical padding operations (repetition or zero-value filling) with a motion compensation-based prediction mechanism. Instead of mechanically copying boundary samples or filling with zeros, the invention substitutes these operations with a more sophisticated process that calculates predicted sample values based on motion vectors and neighboring block content. This substitution improves prediction quality while maintaining computational efficiency through the use of existing motion compensation algorithms.
Solution Approach 2:
The patent applies parameter changes by dynamically selecting different padding strategies based on the characteristics of the current block and available reference samples. The method adjusts the padding approach (using different motion vectors, different neighboring blocks, or different prediction modes) according to the specific coding context, thereby optimizing both prediction quality and computational efficiency for different scenarios rather than using a fixed simple padding method.
3Manufacturing precision
If multiple reference blocks are used for bi-directional prediction, then the prediction accuracy improves, but the complexity of handling out-of-boundary cases increases
Solution Approach 1:
The patent applies segmentation by dividing the bi-directional prediction process into separate handling for each reference list (List 0 and List 1). The padding operation is performed independently for each reference picture, allowing the complexity of handling out-of-boundary cases to be managed separately for each direction. This segmentation enables the encoder to process each reference list with its own motion vectors and padding requirements, simplifying the overall control logic while maintaining high prediction accuracy through the combination of multiple reference blocks.
Data Source
AI summary
A method for encoding a video picture into a bitstream of encoded video picture data, includes: temporal predicting a video picture block by obtaining a temporal predicted block based on at least one reference block of at least one reference picture pointed to by at least one motion vector associated with the video picture block, wherein if the temporal prediction of the video picture block is based on a single reference block of a reference picture, the method further includes padding the reference block based on a motion vector derived from a nearest subblock of the current block inside the video picture.