Video Encoding Transform Selection via Partition Shape Segmentation
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
In video encoding systems, the increased complexity of selecting partition sizes and transform sizes for high-definition videos leads to a higher amount of additional information required for decoding, making it difficult to concentrate energy on fewer transform coefficients, especially when the difference between large and small transform sizes is significant, resulting in wasted additional information.
Innovation Solution
A video encoding apparatus that determines a partition structure, produces predictive images, applies transforms, and variable-length-codes transform selection flags based on a list of transform candidates and prohibited transforms, using partition shape information to limit the selection of transform sizes to highly effective ones, thereby reducing the amount of additional information required.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Adaptability or versatility
If multiple transform sizes are selected for high-definition videos to adapt to local video properties, then the adaptability to local video properties is improved, but the amount of additional information required for decoding increases
Solution Approach 1:
The transform size selection is segmented into multiple categories based on partition shape characteristics (e.g., square partitions, wide partitions, tall partitions). Each category has a predefined list of candidate transform sizes, avoiding the need to transmit full selection information for every block. This segmentation reduces the additional information required while maintaining adaptability to different local video properties.
Solution Approach 2:
Different transform size candidates are assigned to different partition shapes based on their local characteristics. For example, square partitions may use square transform sizes, while wide partitions use wide transform sizes. This local quality approach ensures that each partition uses the most appropriate transform size for its specific shape, improving adaptability without requiring excessive additional information.
2Loss of energy
If the difference between large and small transform sizes is significant, then the energy concentration on fewer transform coefficients is improved, but the amount of additional information required increases and may be wasted
Solution Approach 1:
The patent changes the parameter of transform size selection by establishing predetermined relationships between partition shapes and candidate transform sizes. Instead of allowing arbitrary selection of significantly different transform sizes, the system uses parameter changes based on partition characteristics to determine appropriate transform sizes, reducing the additional information required while maintaining effective energy concentration.
Solution Approach 2:
The patent applies partial action by providing a limited set of candidate transform sizes for each partition shape category rather than allowing full flexibility. This partial selection approach is sufficient to achieve effective energy concentration for each partition type without transmitting excessive additional information, balancing compression efficiency with bitrate constraints.
Data Source
AI summary
While maintaining a high degree of freedom in choosing partition sizes and transformation sizes adapted for local characteristics of videos, the amount of metadata is decreased. A video encoding apparatus (10) divides an input video into blocks of a prescribed size and encodes the video block by block. The video encoding apparatus is provided with: a prediction parameter determining portion (102) that decides the block partition structure; a predictive image producing portion (103) that generates predictive images, partition by partition, as prescribed by the partition structure; a transform coefficient producing portion (107) which applies one of the frequency transformations included in a prescribed transformation preset to prediction residuals, i.e. the differences between predictive images and the input video; a transform restriction deriving portion (104) which generates the list of transform candidate, i.e. lists of frequency transformations that can be applied to each partition, on the basis of partition format information; and a variable-length-encoding portion (108) which, on the basis of the list of transform candidate and the transformation preset, performs variable-length encoding on transformation selection flags.