Bit-Depth-Aware Transform Coefficient Scaling for Image Coding
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing image coding schemes face inefficiencies in expressing large residuals due to increased truncSuffixLen in EG(k) coding, leading to decreased precision and increased complexity when bit-depth is increased, as maintaining a fixed total value for the length of prefix, exp, and escape portions of the EG(k) code limits the range of transform coefficients.
Innovation Solution
An image decoding apparatus with a TU decoder that decodes syntax elements for transform coefficients, using a scaling unit to apply quantization-parameter dependent values and clipping within defined ranges, and an inverse conversion unit to process intermediate coefficients, adjusting shift values based on bit-depth to optimize precision and efficiency.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If truncSuffixLen is increased in accordance with bit-depth to allow large residuals to be accurately expressed, then measurement precision is improved, but device complexity increases
Solution Approach 1:
The patent changes the parameter truncSuffixLen from being dynamically adjusted according to bit-depth to being fixed at a predetermined value. This parameter change resolves the contradiction by eliminating the complexity increase while maintaining sufficient precision through the fixed value, which is optimized to handle the majority of cases without requiring dynamic adjustment.
Solution Approach 2:
The patent applies partial action by using a fixed truncSuffixLen value that is sufficient for most practical cases rather than dynamically adjusting it to cover all possible bit-depth scenarios. This partial approach achieves adequate precision for typical applications while avoiding the complexity of dynamic parameter adjustment.
2Measurement precision
If truncSuffixLen is increased in accordance with bit-depth, then the range of transform coefficients is extended, but the length of maxPreExtLen decreases to maintain fixed total value, preventing efficient expression of large residual values
Solution Approach 1:
The patent changes the parameter truncSuffixLen from a dynamic value that decreases with increased bit-depth to a fixed predetermined value. This parameter change resolves the contradiction by allowing both a sufficient transform coefficient range and efficient large residual expression, as the fixed value is optimized to maintain balance between these requirements across different bit-depth scenarios.
3Measurement precision
If the escape portion of EG(k) code is made larger than necessary, then measurement precision is improved, but device complexity increases
Solution Approach 1:
The patent changes the escape portion size from a dynamically adjustable parameter to a fixed predetermined value. This parameter change resolves the contradiction by eliminating the need for complex dynamic adjustment mechanisms while maintaining sufficient precision for accurate coefficient representation through the optimized fixed value.
Data Source
AI summary
Provided are an image decoding apparatus (31) and an image coding apparatus (11) in which both transform precision and residual coding efficiency are preferable. Included are a TU decoder (3024) configured to decode, from coded data, a syntax element representing an absolute value of a transform coefficient, a scaling unit (311) configured to multiply a transform coefficient by a quantization-parameter dependent value and perform right-shift by a first shift value, and to perform clipping using a minimum value and a maximum value according to a first range, and an inverse conversion processing unit (311) configured to convert a scaled transform coefficient, perform clipping using the minimum value and the maximum value according to the first range to derive an intermediate coefficient, further transform an intermediate variable, and perform right-shift by a second shift value. The scaling unit (311) derives the first range, the first shift value, and the second shift value in accordance with a value of a first bit-depth, and limits the first range to be equal to or less than a prescribed value.