Palette Coding Non-Local Prediction Video Compression
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Conventional palette coding in HEVC systems struggles to effectively utilize non-local duplicated contents for improved coding efficiency, particularly in screen content coding, as existing modes like COPY_ABOVE cannot handle non-local duplicated indexes effectively.
Innovation Solution
The introduction of a non-local prediction (NLP) mode in palette coding, which allows for the use of motion vector information to locate a reference block from within or outside the current picture, enabling the prediction of current block samples or indices by corresponding reference samples or indices, and allowing transformations such as flipping, rotation, and transposition of the reference block to generate a non-local index map.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Productivity
If COPY_ABOVE mode is used for palette coding, then local spatial redundancy is exploited, but non-local duplicated contents cannot be utilized
Solution Approach 1:
The prediction mode is segmented into multiple types: COPY_ABOVE for local vertical prediction, COPY_LEFT for local horizontal prediction, and COPY_INTER for non-local prediction. Each segment handles specific prediction scenarios, allowing the system to exploit both local and non-local redundancies effectively.
Solution Approach 2:
The invention extends the prediction from two-dimensional local spatial neighbors (above and left) to include temporal dimension by introducing inter-picture prediction. Motion vectors enable prediction from reference pictures in previous or future frames, adding a temporal dimension to the prediction process.
2Device complexity
If only local prediction modes (COPY_ABOVE, COPY_LEFT) are used, then implementation is simple, but compression performance for non-local duplicated content is poor
Solution Approach 1:
The prediction mode selection is made dynamic rather than static. The encoder chooses between COPY_ABOVE, COPY_LEFT, and COPY_INTER modes based on the actual content characteristics and redundancy patterns. This dynamic adaptation allows the system to achieve better compression performance while maintaining reasonable complexity through selective application of each mode.
Solution Approach 2:
The invention uses copying mechanisms at multiple levels: copying palette indices from above row (COPY_ABOVE), from left column (COPY_LEFT), and from reference pictures using motion compensation (COPY_INTER). This multi-level copying approach efficiently exploits redundancy while keeping the implementation relatively simple.
3Productivity
If inter-picture prediction is introduced, then non-local correlations are exploited, but additional motion vector information must be signaled
Solution Approach 1:
Motion vector information is copied from the corresponding Inter-prediction 2N×2N coding unit rather than being fully re-encoded. This copying approach reduces the bitstream overhead significantly while still enabling effective non-local prediction, as the motion vectors are already available from the co-located inter-coded unit.
Solution Approach 2:
The motion vector information serves multiple functions: it enables inter-picture prediction for palette mode, maintains temporal consistency with co-located inter-coded units, and reduces bitstream overhead through reuse. This multi-functionality justifies the inclusion of motion vector signaling in the bitstream.
Data Source
AI summary
Method and apparatus for video coding using palette coding modes including a copy inter mode are disclosed. When the copy inter mode is selected, one or more consecutive current samples or indices of the current block are predicted by one or more consecutive reference samples or indices of a reference block at corresponding locations. Motion vector (MV) information for locating the reference block can be signalled in a bitstream for the current block. The motion vector information may be related to MV for the reference block corresponding to an Inter-prediction coding unit (CU) or a block vector (BV) for an Intra Block Copy (IntraBC) coded block. The location of the reference block can be determined implicitly without position information of the reference block signalled in a bitstream for the current block.