Quantization Method for Concurrent RD Cost Calculation
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
In rate distortion optimized quantization (RDOQ) processing for HEVC, the dependency among coefficients leads to sequential processing, which hinders concurrent operation and increases processing time, affecting the tradeoff between distortion and rate in coding.
Innovation Solution
A quantization method that performs scaling on transform coefficients, calculates RD costs based on quantization error and code amounts, and determines coefficient absolute values and positions to optimize RD cost, allowing concurrent processing by removing dependency relations among coefficients.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If sequential processing is used to calculate RD costs for each coefficient, then dependency among coefficients is maintained ensuring accurate rate distortion optimization, but processing time increases and productivity decreases
Solution Approach 1:
The patent divides the target block into multiple sub-blocks and processes coefficients in parallel within each sub-block. By segmenting the processing domain and establishing independence within segments, the patent enables concurrent RD cost calculation while maintaining accuracy through proper segmentation boundaries.
Solution Approach 2:
The patent performs preliminary classification of coefficients into different groups based on their properties before RD cost calculation. This preliminary action groups coefficients that can be processed independently, enabling parallel processing while ensuring that each group maintains the necessary dependency relationships for accurate RD optimization.
2Productivity
If concurrent processing is implemented by removing dependency relations among coefficients, then processing time decreases and productivity improves, but coding efficiency may deteriorate due to suboptimal RD cost calculation
Solution Approach 1:
By segmenting coefficients into independent groups where intra-group dependencies are preserved but inter-group dependencies are removed, the patent enables parallel processing. Each segment processes coefficients with their local dependencies intact, preventing information loss while achieving concurrency through inter-segment independence.
Solution Approach 2:
The patent applies different processing strategies to different coefficient groups based on their local characteristics. By making the processing quality local rather than uniform, the patent maintains high RD optimization accuracy within each group while enabling global parallel execution across independent groups.
Data Source
AI summary
A quantization method includes: performing, by a computer, scaling on transform coefficients of a target block; quantizing scaled transform coefficients to obtain a plurality of quantization coefficients; calculating, when calculating, based on a quantization error and a first code amount of a coded target block obtained by coding the target block, a cost of coding the target block by using a coefficient absolute value candidate of a plurality of coefficient absolute value candidates of one of the plurality of quantization coefficients, the first code amount according to a second code amount of a coded element for the one of the plurality of quantization coefficients, the second code amount being obtained by binarizing the coded element under one of a first condition and a second condition.