CCLM Prediction Parameter Derivation Using Right-Shift Scaling
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
The existing CCLM prediction methods require large memory for table storage and suffer from complexity due to high bit width products, particularly in deriving linear prediction parameters and generating prediction images.
Innovation Solution
A CCLM prediction unit that derives CCLM prediction parameters (a, b) using luma and chroma difference values and a table, and generates a chroma prediction image by shifting the product of parameter a and luma by a prescribed shift value, thereby simplifying multiplication and reducing memory requirements.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If the existing CCLM prediction method uses a table for deriving linear prediction parameters, then the prediction accuracy is improved, but the memory usage increases significantly
Solution Approach 1:
The patent changes the representation parameters of the linear prediction parameters a and b. Instead of storing full-precision values in a large table, the patent represents a as a fraction with denominator 2^shiftA and b as an integer, enabling compact storage while maintaining prediction accuracy through controlled precision loss via right-shift operations.
2Measurement precision
If the bit width of CCLM prediction parameter a is increased to improve precision, then the prediction accuracy is improved, but the complexity of multiplication operations increases
Solution Approach 1:
The patent changes the bit width and representation of parameter a by expressing it as a fraction with power-of-2 denominator (2^shiftA). This allows the multiplication a*luma to be simplified to (a_num*luma)>>shiftA, where the division by power of 2 is replaced by a simple right-shift operation, significantly reducing hardware complexity while maintaining adjustable precision through the shiftA parameter.
3Measurement precision
If the existing method uses products of large bit width values for deriving prediction parameters, then the prediction accuracy is improved, but the hardware scale increases
Solution Approach 1:
The patent changes the numerical representation of parameters a and b such that a is expressed as a fraction with power-of-2 denominator and b as an integer. This transformation allows the complex multiplication operations (a*luma and a*Y_MIN) to be replaced by simpler operations involving multiplication followed by right-shift, reducing the bit width requirements and hardware scale while maintaining sufficient precision through the shift parameters.
Data Source
AI summary
The amount of memory required for CCLM prediction is reduced. The CCLM prediction parameter derivation unit (310442) derives a scale shift value corresponding to a luma difference value, and derives a CCLM prediction parameter, by shifting, by using the scale shift value, a value obtained by multiplying a value of a table referred to with a value obtained by performing right shift of the luma difference value by the scale shift value as an index and a chroma difference value. In addition, in a case of deriving a prediction image, a bit width is reduced by adaptively deriving a shift amount of a linear prediction parameter from the chroma difference value.