High-Precision DCT Transform Matrices for Video Compression

Resolve Bottlenecks,
Find Innovative Solutions
Generate Solutions

Solution Overview

Problem

Existing video data encoding and decoding systems face inefficiencies in compressing and decompressing video data, particularly with high bit depth, due to limitations in transform matrix precision and entropy encoding techniques.

Innovation Solution

Implementing forward transform matrices with higher precision, specifically 14-bit coefficients, matched to standard inverse matrices, for video data encoding and decoding processes, using discrete cosine transforms (DCT) and discrete sine transforms (DST) to enhance compression efficiency.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Productivity

If standard precision transform matrices are used for video encoding, then device complexity is reduced, but compression efficiency deteriorates for high bit depth video data

Engineering Contradiction:
Improvecompression efficiencyVSAvoidtransform matrix precision
Core Design Contradiction:
ProductivityVSDevice complexity

Solution Approach 1:

The patent applies parameter changes by increasing the precision of transform matrix coefficients from standard precision to 14-bit precision. This parameter change directly improves compression efficiency for high bit depth video data by providing more accurate frequency domain representation, while the increased precision is managed through efficient memory structures and processing techniques that mitigate the complexity increase.

Inventive Principle:
Principle #35Parameter changes

2Productivity

If higher precision transform matrices are used, then compression efficiency improves, but computational complexity increases

Engineering Contradiction:
Improvecompression efficiencyVSAvoidcomputational complexity
Core Design Contradiction:
ProductivityVSPower

Solution Approach 1:

The patent changes the precision parameter of transform matrices to 14-bit, which improves compression efficiency by enabling more accurate transformation of video data into frequency domain coefficients. The computational complexity increase is addressed through optimized algorithms and data structures that handle the higher precision requirements efficiently.

Inventive Principle:
Principle #35Parameter changes

3Measurement precision

If 14-bit transform matrix coefficients are used, then the accuracy of frequency domain representation improves, but memory requirements increase

Engineering Contradiction:
Improvefrequency domain representation accuracyVSAvoidmemory requirements
Core Design Contradiction:
Measurement precisionVSQuantity of substance

Solution Approach 1:

The patent increases the precision parameter of transform matrix coefficients to 14-bit, which significantly improves the accuracy of frequency domain representation for high bit depth video data. The increased memory requirements are managed through efficient memory allocation strategies and data structures optimized for the specific precision requirements of the transform operations.

Inventive Principle:
Principle #35Parameter changes

Data Source

PatentEP4128768B1Video data encoding and decoding
Publication Date: 2025.11.26 SONY GROUP CORP
  • EP4128768B1 patent drawingFigure 1~6
  • EP4128768B1 patent drawingFigure 7
  • EP4128768B1 patent drawingFigure 8

AI summary

A video data encoding method for encoding an array of video data values comprises the steps of: frequency-transforming the video data values according to a frequency transform, to generate an array of frequency-transformed values by a matrix-multiplication process using a transform matrix having a data precision of fourteen bits, the frequency transform being a discrete cosine transform; defining a 64 x 64 transform matrix M64 for a 64 x 64 DCT transform, the matrix M64 being defined by the accompanying Figures 10a- 10e and 11; for an N x N transform where N is 2, 4, 8 or 16, subsampling the 64 x 64 transform matrix Me4 to select a subset of N x N values, the subset ΜΝ[x][y] of values being defined by: M N [x][y] = M 64 [x][(2 6-log2(N) ))y] for x, y = 0..(N- 1).