Motion Vector Difference Entropy Coding With Fewer Contexts

The patent segments the binarization process into two distinct parts: truncated unary code for small motion vector differences (0-2) and exponential Golomb code for larger values (3+). This segmentation allows each coding method to be optimized independently, with the truncated unary code using a single context model that avoids the complexity of multiple contexts while maintaining coding precision for the most frequent small values.

The patent changes the cutoff parameter from the conventional value of 9 to 2, which fundamentally alters the coding strategy. By setting cutoff=2, the truncated unary code handles only the two most significant bins (0 and 1) with a single context, while the exponential Golomb code handles the remaining bins. This parameter change reduces context complexity while maintaining or improving coding efficiency.

The patent segments the context modeling into a single unified context for the truncated unary code bins, rather than providing multiple contexts for each bin position. This segmentation simplifies probability adaptation by ensuring sufficient sampling for each context, allowing the probability estimation to converge reliably and perform effective probability adaptation.

By changing the cutoff parameter to 2, the patent ensures that the single context for truncated unary code bins receives sufficient probability adaptation samples. This parameter change resolves the probability adaptation failure by balancing the number of bins per context with the available training data, enabling reliable probability estimation.

The patent segments the coding approach by applying different code types to different ranges of motion vector difference values. The truncated unary code with a single context handles bins 0 and 1, while exponential Golomb code handles bins 3 and above. This segmentation allows the system to maintain low device complexity while achieving high coding efficiency by matching the coding method to the statistical characteristics of each bin range.

By setting cutoff=2, the patent optimizes the boundary between the two coding methods. This parameter change ensures that bins with highly varying symbol statistics are appropriately handled: bins 0 and 1 use the simplified single-context truncated unary code, while bins 3+ use the more flexible exponential Golomb code, thereby maintaining coding efficiency without requiring multiple contexts for each bin.

The patent segments the context requirements by applying a single context model to the truncated unary code bins (0 and 1), eliminating the need to inspect neighboring bins for context selection. This segmentation significantly reduces decoding execution time for the most frequent small motion vector differences, while exponential Golomb code bins (3+) can use their own context mechanisms if needed.

By changing the cutoff to 2, the patent reduces the number of bins requiring complex multi-context inspection. Only bins 0 and 1 use the single-context truncated unary code, minimizing the time loss from context inspection. This parameter change optimizes the trade-off between coding precision and decoding speed by applying the simplified single-context approach to the most frequently occurring bins.