Non-uniform Reconstruction Space for Video Compression
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current video encoding standards, such as H.264, use uniform quantization step sizes, which do not always achieve optimal rate-distortion performance due to the assumption of a uniform data distribution, leading to suboptimal compression efficiency.
Innovation Solution
Implementing a non-uniform reconstruction space in data compression methods and devices, where the encoder determines and communicates non-uniform reconstruction parameters to the decoder for dequantization, allowing for adaptive reconstruction levels based on actual data distribution, thereby improving rate-distortion performance.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Manufacturing precision
If uniform quantization step size is used, then the encoding process is simple and consistent, but the rate-distortion performance is suboptimal due to mismatch with non-uniform data distribution
Solution Approach 1:
The patent applies local quality by using non-uniform reconstruction spaces that are adapted to the local statistical properties of the data. Different regions of the data space use different reconstruction characteristics matched to their local distribution patterns, thereby improving rate-distortion performance without requiring complete redesign of the entire encoding system.
Solution Approach 2:
The patent changes the reconstruction parameter from a uniform step size to a non-uniform distribution that matches the actual data statistics. By adjusting the reconstruction space parameters to reflect the true data distribution, the system achieves better compression efficiency and rate-distortion performance.
2Productivity
If non-uniform reconstruction space is implemented, then the rate-distortion performance is improved, but the encoder and decoder complexity increases
Solution Approach 1:
The patent applies preliminary action by pre-determining the non-uniform reconstruction space parameters at the encoder side based on data statistics. These parameters are then communicated to the decoder, allowing both sides to use the optimized reconstruction space without real-time computation during the actual encoding/decoding process, thus limiting the increase in complexity.
Solution Approach 2:
The patent introduces an intermediary element - the non-uniform reconstruction space parameters - that mediates between the uniform quantization process and the final reconstruction. These parameters act as a bridge that enables improved compression efficiency while maintaining a relatively simple encoder-decoder structure.
3Measurement precision
If uniform partitioning of data space is used, then the quantization process is straightforward, but the reconstruction does not match actual data distribution
Solution Approach 1:
The patent applies local quality by making the reconstruction space properties match the local data distribution characteristics. Different regions use reconstruction parameters appropriate to their local statistics, improving data representation accuracy while keeping the quantization process relatively simple through pre-computed parameters.
Solution Approach 2:
The patent essentially creates a copied or modeled version of the data distribution as the reconstruction space. Instead of directly using the complex actual data distribution, a simplified model (the non-uniform reconstruction space) is created that captures the essential statistical properties, achieving good representation accuracy with manageable complexity.
Data Source
Figure 1
Figure 2
Figure 3~4
AI summary
Encoding and decoding methods are presented that use adaptive reconstruction levels. Reconstruction space parameters are developed by an encoder and inserted in the bitstream with the encoded video data. The reconstruction space parameter may include parameters from which the decoder can determine the levels for dequantization of the encoded video data. The reconstruction space parameters may include a first reconstruction level and a step size between other levels. The first reconstruction level may not equal the step size. In some cases, neither may be equal to the quantization step size used to quantize the transform domain coefficients.