Image Transform Coding With NSPT and LFNST for Block Efficiency

Resolve Bottlenecks,
Find Innovative Solutions
Generate Solutions

Solution Overview

Problem

Existing image compression technologies face challenges in efficiently encoding and decoding high-resolution and high-quality images, particularly in handling non-separable transforms and determining optimal transform kernels for improved encoding efficiency.

Innovation Solution

The method employs a non-separable primary transform (NSPT) and low frequency non-separable transform (LFNST) to derive and encode transform coefficients, with the width and height of the transform region determined based on the current block dimensions, applying specific output lengths based on block sizes for enhanced encoding and decoding processes.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Productivity

If non-separable primary transform (NSPT) is used for high-resolution image compression, then transform performance and encoding efficiency are improved, but device complexity and computational load increase

Engineering Contradiction:
Improveencoding efficiencyVSAvoiddevice complexity
Core Design Contradiction:
ProductivityVSDevice complexity

Solution Approach 1:

The patent divides the transform process into two stages: first applying a standard separable transform (DCT/DST) to all blocks, then selectively applying NSPT only to specific blocks based on prediction mode and block characteristics. This segmentation allows the system to gain NSPT performance benefits for relevant blocks while avoiding the computational overhead of applying NSPT to all blocks, thus resolving the contradiction between encoding efficiency and device complexity.

Inventive Principle:
Principle #1Segmentation

Solution Approach 2:

The patent applies different transform methods to different regions/blocks based on local characteristics such as intra prediction mode, block size, and gradient information. Blocks with specific characteristics (e.g., certain intra prediction modes, specific size ranges) receive NSPT treatment while other blocks use standard transforms. This local quality approach ensures optimal transform performance is applied where needed without unnecessarily increasing complexity everywhere.

Inventive Principle:
Principle #3Local quality

2Manufacturing precision

If NSPT is applied to all blocks, then transform performance is maximized, but processing time and computational resources increase significantly

Engineering Contradiction:
Improvetransform performanceVSAvoidprocessing time
Core Design Contradiction:
Manufacturing precisionVSLoss of time

Solution Approach 1:

The patent introduces dynamic decision-making mechanisms that adaptively select whether to apply NSPT to each block based on real-time characteristics such as prediction mode, block size, and local image content. This dynamic approach allows the system to maximize transform performance for blocks that benefit from NSPT while minimizing processing time by using faster standard transforms for blocks where NSPT provides little benefit, thus resolving the contradiction between transform performance and processing time.

Inventive Principle:
Principle #15Dynamics

3Manufacturing precision

If transform kernels are optimized for specific block sizes, then encoding precision is improved, but adaptability to various block sizes decreases

Engineering Contradiction:
Improveencoding precisionVSAvoidadaptability to block sizes
Core Design Contradiction:
Manufacturing precisionVSAdaptability or versatility

Solution Approach 1:

The patent employs multiple NSPT kernels with different dimensions (e.g., 4x4, 8x8, 16x16) that can be selected based on block size. The system includes logic to match the appropriate kernel size to the block being transformed, and also handles cases where blocks are divided into sub-blocks to apply smaller kernels. This parameter change approach allows the system to maintain high encoding precision for each block size while preserving adaptability across various block dimensions.

Inventive Principle:
Principle #35Parameter changes

Data Source

PatentEP4697705A1Image encoding/decoding method and device, and recording medium on which bitstream is stored
Publication Date: 2026.02.18 LG ELECTRONICS INC
  • EP4697705A1 patent drawingFigure 1
  • EP4697705A1 patent drawingFigure 2
  • EP4697705A1 patent drawingFigure 3

AI summary

An image decoding method and device according to the present disclosure can acquire residual information from a bitstream, derive transform coefficients of the current block on the basis of the residual information, derive residual samples of the current block by performing inverse quantization and/or inverse transform on the transform coefficients of the current block, and reconstruct the current block on the basis of the residual samples of the current block. Here, the inverse transform can be performed on the basis of a non-separable primary transform and/or a low frequency non-separable transform.