Piece-wise linear mapping for in-loop reshaping
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current video coding standards face challenges in efficiently reshaping and coding images, particularly with higher bit depths and dynamic range, which affects compression efficiency and image quality, especially in next-generation video coding standards like VVC.
Innovation Solution
The implementation of piece-wise linear (PWL) representation methods for reshaping functions in video coding, where a processor generates and adjusts PWL representations of forward reshaping functions to map luma pixel values across different codeword representations, and decoders use pivot parameters to adjust these representations for accurate decoding, enabling improved compression efficiency and image quality.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If piece-wise linear representation with more segments (e.g., 16 segments) is used for reshaping functions, then mapping precision and image quality are improved, but computational complexity and processing overhead increase
Solution Approach 1:
The patent applies segmentation by dividing the reshaping function into multiple piece-wise linear segments. Each segment represents a portion of the dynamic range with its own linear mapping parameters, allowing precise control over different regions of the codeword distribution while maintaining computational efficiency through localized processing rather than a single complex non-linear function.
Solution Approach 2:
The patent changes parameters by representing the reshaping function through a set of linear parameters (slope and intercept for each segment) instead of a complex non-linear function. This parameter transformation allows the system to achieve high mapping precision through simple linear operations that can be efficiently implemented in hardware and software, reducing computational complexity while maintaining accuracy.
2Device complexity
If piece-wise linear representation with fewer segments (e.g., 8 segments) is used for reshaping functions, then computational complexity is reduced, but mapping precision and image quality deteriorate
Solution Approach 1:
The patent applies partial action by using a reduced number of segments (e.g., 8 instead of 16) where sufficient precision is achieved for the specific application requirements. This selective reduction eliminates unnecessary computational overhead while maintaining adequate mapping precision, demonstrating that full segmentation is not always required and allowing optimization based on specific performance needs.
3Measurement precision
If non-linear opto-electronic functions (OETF) are used for quantization, then image quality is improved through perceptual encoding, but compression efficiency decreases due to uneven codeword distribution
Solution Approach 1:
The patent applies inversion by reversing the traditional approach: instead of applying non-linear OETF and then dealing with uneven codeword distribution, it directly applies piece-wise linear mapping functions that are designed to create uniform or optimized codeword distribution from the start. This inverted approach eliminates the need for separate non-linear quantization and redistribution steps, simultaneously achieving perceptual quality and compression efficiency.
Solution Approach 2:
The patent changes parameters by transforming the codeword distribution characteristics through piece-wise linear mapping. By adjusting the slope and intercept parameters of each linear segment, the system can control the density and distribution of codewords to match perceptual importance and statistical properties of the image data, achieving both quality preservation and compression optimization in a unified transformation.
Data Source
AI summary
Given a representation of a forward reshaping function in an M-piecewise linear representation (M-PWL), methods, processes, and systems are presented for generating the forward reshaping function in an N-PWL representation (where N<M), by properly adjusting the first and last segments, and for generating a decoded bitstream using the adjusted N-PWL reshaping representation. Syntax methods for signaling reshaping parameters are also presented.