Unified Linear Model Estimation for Video Coding
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current video coding systems, particularly those employing template-based coding techniques like Cross Component Linear Model (CCLM) and Local Illumination Compensation (LIC), face challenges due to burdensome hardware implementations and individual and combined computation complexities.
Innovation Solution
The proposed solution involves an improved linear model estimation approach for template-based video coding techniques, which uses a unified linear model parameter estimation method. This method computes slope and intercept parameters using a single look-up table (LUT) at a precision no greater than 16 bits, simplifying the computation and reducing memory requirements.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Productivity
If template-based coding techniques (CCLM and LIC) are used to improve video coding efficiency, then compression performance is improved, but hardware implementation complexity and computation complexity increase
Solution Approach 1:
The patent combines CCLM and LIC into a unified template-based coding framework that shares common computational components, particularly the linear model parameter estimation mechanism. This merging reduces redundant hardware circuits and simplifies the overall implementation while maintaining the compression benefits of both techniques.
Solution Approach 2:
The patent creates a universal linear model estimation apparatus that can perform both CCLM chroma prediction and LIC illumination compensation functions. This multi-functional design eliminates the need for separate dedicated hardware for each technique, reducing device complexity while preserving coding efficiency.
2Productivity
If template-based coding techniques (CCLM and LIC) are used to improve video coding efficiency, then compression performance is improved, but computation complexity increases
Solution Approach 1:
The patent merges the parameter estimation computations of CCLM and LIC into a single unified process. By sharing the linear model calculation infrastructure, the patent reduces the total computational operations required while maintaining the predictive accuracy benefits of both techniques.
Solution Approach 2:
The patent optimizes computational parameters by limiting precision to 16-bit fixed-point arithmetic and using pre-computed lookup tables for linear model parameters. This parameter optimization reduces the computational burden while maintaining sufficient accuracy for video coding applications.
3Measurement precision
If high precision linear model parameter estimation is used to improve prediction accuracy, then coding quality is improved, but memory requirements and computation complexity increase
Solution Approach 1:
The patent changes the precision parameter from high-precision floating-point to 16-bit fixed-point arithmetic, and implements lookup tables with optimized entry counts. This parameter optimization achieves a balance between memory consumption and estimation accuracy suitable for video coding applications.
Solution Approach 2:
The patent uses simplified 16-bit fixed-point calculations instead of high-precision floating-point arithmetic, accepting a small loss in precision in exchange for dramatically reduced memory requirements and computational complexity. This approach is sufficient for video coding where absolute precision is less critical than overall compression efficiency.
Data Source
AI summary
Methods and apparatuses for encoding and decoding video data. Techniques disclosed comprise obtaining a linear model used to encode or to decode a video block of the video data. The linear model is obtained by computing a parameter of the linear model as a function of a minimum chroma value and a maximum chroma value and as a function of a reciprocal of a difference between a minimum luma value and a maximum luma value, wherein the reciprocal is derived from a single look up table including values for determining least significant bits.