Inverse DCT Pruning via Group Significance Map
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current video coding standards, such as H.264/AVC and HEVC, face high computational complexity in inverse discrete cosine transformation (IDCT) due to large transform sizes, leading to inefficient pruning techniques that do not accurately determine the number of non-zero coefficients, resulting in unnecessary computations.
Innovation Solution
The method employs a group significance map to identify non-zero transform coefficients in a transform block, allowing for optimized one-dimensional IDCT pruning by determining the exact region of non-zero coefficients, reducing computational complexity by skipping zero-value computations, and applying IDCT in opposite directions (vertical and horizontal) to generate a residual block.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Productivity
If traditional IDCT pruning techniques are used in video coding standards, then some computational complexity is reduced, but the pruning is inefficient and does not accurately determine the number of non-zero coefficients, resulting in unnecessary computations
Solution Approach 1:
The patent applies preliminary action by using a group significance map to identify and mark non-zero transform coefficients before performing the IDCT operation. This allows the decoder to pre-determine the exact region containing non-zero coefficients, enabling subsequent pruning of zero-value computations without sacrificing accuracy. The significance map is generated during encoding and transmitted to the decoder, where it guides the IDCT process to skip unnecessary computations on zero-coefficient regions.
2Loss of information
If the transform block size is increased to improve compression ratio, then more information is preserved, but the computational complexity of IDCT increases significantly
Solution Approach 1:
The patent applies the extraction principle by isolating and identifying only the non-zero coefficients within the transform block using the group significance map. Instead of processing the entire large transform block, the IDCT operation is extracted and applied only to the specific region containing non-zero coefficients. This dramatically reduces computational complexity for large transform blocks while preserving all necessary information, as zero-coefficient regions are excluded from computation.
3Device complexity
If conventional pruning methods are applied, then some zero computations are skipped, but the bound on non-zero coefficients is loose and accuracy is compromised
Solution Approach 1:
The patent replaces the conventional mechanical pruning approach with a signal-processing-based solution using the group significance map. Instead of relying on loose bounds or heuristic methods to estimate the region of non-zero coefficients, the significance map provides precise, binary identification of which coefficients are non-zero. This substitution of the identification mechanism achieves both computational efficiency and high accuracy, as the map is generated based on actual coefficient values during encoding.
Data Source
AI summary
A method for inverse discrete cosine transformation (IDCT) in video coding is provided that includes receiving a transform block, identifying a region of non-zero transform coefficients in the transform block using a group significance map corresponding to the transform block, wherein any transform coefficients not in the region have a value of zero, applying a one-dimensional (1D) IDCT to the region of non-zero transform coefficients in a first direction to generate an interim results block, wherein 1D IDCT computations are not performed on transform coefficients outside the region, and applying a 1D IDCT to the interim results block in a second direction to generate a residual block.