Motion Vector Prediction Using Spatial Temporal Candidates
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
In image compression methods like MPEG and H.264, the encoding of motion vectors increases data overhead, reducing the image compression ratio, and existing motion vector prediction methods are not effective when neighboring blocks have different motion vectors.
Innovation Solution
An apparatus that decodes motion vectors using spatial and temporal motion vector candidates, where the temporal merge candidate has a reference picture index of 0 and a motion vector from a corresponding picture, and scales motion vectors when reference pictures differ, to efficiently encode motion information.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If motion vector information is encoded and inserted as it is, then the motion vector can be accurately decoded, but the compression ratio of image data is decreased because overhead is increased
Solution Approach 1:
The patent creates a copy of motion vector information from neighboring blocks and reference pictures. Instead of encoding the actual motion vector, the encoder transmits an index pointing to a candidate in the motion vector candidate list, which is constructed by copying motion vectors from spatially or temporally neighboring blocks. This significantly reduces the number of bits required to represent motion vector information while maintaining accurate motion compensation.
2Ease of operation
If motion vector prediction uses median of neighboring blocks, then prediction can be performed, but the prediction is not effective when neighboring blocks have different motion vectors
Solution Approach 1:
The patent implements a dynamic motion vector prediction system where the prediction method adapts based on the characteristics of neighboring blocks. The system constructs a candidate list that includes multiple types of predictions (spatial neighboring blocks, temporal reference pictures, and zero vectors) and selects the best match dynamically. This allows the system to switch between different prediction strategies depending on the scene content and motion patterns, improving prediction accuracy for various motion scenarios.
3Measurement precision
If more motion vector candidates are considered, then prediction accuracy improves, but computational complexity increases
Solution Approach 1:
The patent applies local quality by constructing motion vector candidate lists with different priorities and weights for different candidate types. Spatial neighboring blocks and temporal reference pictures are assigned different positions in the candidate list, allowing the decoder to try candidates in order of likelihood without exhaustively evaluating all possibilities. This localized prioritization reduces computational complexity while maintaining high prediction accuracy for most cases.
Data Source
AI summary
Provided is an apparatus for decoding a moving picture. An inverse quantization/transformation unit generates a quantized block by inversely scanning a quantized coefficient sequence, generates a transform unit by inversely quantizing the quantized block using a quantization step size, and generates a residual block by inversely transforming the transform block. An inter prediction unit generates a prediction block of a current prediction unit based on motion vector information. An adding unit generates a restored block using the residual block and the prediction block. When a motion information coding mode is a skip mode, the inter prediction unit restores motion information of the current prediction block using an available spatial or temporal skip candidate. The temporal skip candidate includes a reference picture index and a motion vector, the reference picture index of the temporal skip candidate is set to 0, and a motion vector of the temporal skip candidate is a motion vector of the temporal skip candidate in a temporal skip candidate picture. A size of the prediction unit is same with a size of a coding unit and the prediction block is set as the restored block.