Selective Probability Distribution Mixing for Video Entropy Coding
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing video compression techniques face challenges in balancing compression performance with the overhead cost associated with context modeling, particularly in systems using multi-symbol, non-binary alphabets.
Innovation Solution
The implementation of selective mixing of probability models allows for the addition of contexts while limiting the overhead, by using a first context model for all tokens and a second context model for more frequent tokens, mixing their probability distributions for encoding.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Productivity
If multiple context models are added to improve compression performance, then compression ratio is improved, but overhead cost increases
Solution Approach 1:
The patent segments the token vocabulary into two groups: frequent tokens and infrequent tokens. Different context models are selectively applied to each group - a first context model handles frequent tokens while a second context model handles infrequent tokens. This segmentation allows the system to use multiple context models to improve compression performance for frequent tokens without incurring excessive overhead for all tokens, thus resolving the contradiction between compression performance and overhead cost.
Solution Approach 2:
The patent applies different quality levels of context modeling to different parts of the token space. For frequent tokens, a more sophisticated second context model is used to achieve better compression. For infrequent tokens, a simpler first context model is used to minimize overhead. This local differentiation in modeling quality allows the system to optimize compression performance where it matters most (frequent tokens) while controlling overall overhead, thereby resolving the technical contradiction.
2Productivity
If context modeling complexity is increased to improve compression efficiency, then compression ratio is improved, but computational complexity increases
Solution Approach 1:
The patent divides the computational workload by segmenting tokens into frequent and infrequent categories. The simpler first context model processes infrequent tokens with lower computational requirements, while the more sophisticated second context model processes frequent tokens where the computational investment yields greater compression returns. This segmentation of computational complexity resolves the contradiction by strategically allocating computational resources.
Solution Approach 2:
The patent applies partial sophisticated context modeling only where necessary - specifically for frequent tokens that contribute most to compression efficiency. Rather than applying complex context modeling uniformly to all tokens (excessive action), the system applies enhanced modeling selectively to the subset of tokens that provide the greatest compression benefit, thus improving compression efficiency without incurring proportional increases in computational complexity.
Data Source
Figure 1
Figure 2
Figure 3
AI summary
An apparatus for decoding transform coefficients using an alphabet of transform coefficient tokens includes a memory and a processor. A first probability distribution corresponding to a first context is selected, and a second probability distribution corresponding to a second context is selected. In response to determining that the second probability distribution includes a probability for a transform coefficient token, the first probability distribution and the second probability distribution are mixed to generate a mixed probability. The transform coefficient token is entropy decoded using the mixed probability. The first probability distribution is defined for all tokens of the alphabet. The second probability distribution is defined over a non-trivial partition of the tokens.