Image Decoding Device Selective BDOF Processing

Resolve Bottlenecks,
Find Innovative Solutions
Generate Solutions

Solution Overview

Problem

Existing techniques for Bi-Directional Optical Flow (BDOF) processing face challenges in reducing processing time consistently across both software and hardware implementations.

Innovation Solution

An image decoding device and method that incorporates a motion vector decoding unit, a refinement unit for correcting motion vectors, and a predictive signal generation unit that determines whether to apply BDOF processing based on information calculated during refinement processing, thereby optimizing processing efficiency.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Loss of time

If BDOF skipping processing is implemented using software, then execution time is shortened, but execution time increases in hardware-based implementation

Engineering Contradiction:
Improveexecution timeVSAvoidprocessing complexity
Core Design Contradiction:
Loss of timeVSDevice complexity

Solution Approach 1:

The patent applies different processing strategies to different blocks based on local characteristics. By calculating sum of absolute differences (SAD) for each block and comparing against a threshold, the system selectively applies BDOF processing only where needed, rather than uniformly across all blocks. This local differentiation resolves the contradiction by optimizing for software execution time while maintaining hardware implementation feasibility.

Inventive Principle:
Principle #3Local quality

Solution Approach 2:

The patent changes the parameter of processing application by introducing a threshold-based decision mechanism. Instead of always applying or never applying BDOF, the system dynamically determines application based on SAD values. This parameter change enables the system to adapt to different implementation contexts (software vs hardware) while maintaining optimal performance.

Inventive Principle:
Principle #35Parameter changes

2Loss of time

If BDOF buffer reduction technique is implemented using hardware, then execution time is shortened, but execution time increases when using software

Engineering Contradiction:
Improveexecution timeVSAvoidprocessing complexity
Core Design Contradiction:
Loss of timeVSDevice complexity

Solution Approach 1:

The patent segments the video processing into distinct stages: motion vector decoding, refinement processing with SAD calculation, and selective BDOF application. By dividing the processing pipeline and allowing independent optimization of each segment for different platforms, the system resolves the contradiction between hardware and software implementation performance.

Inventive Principle:
Principle #1Segmentation

3Productivity

If selective BDOF processing is applied based on refinement processing information, then processing amount is reduced in hardware implementation, but processing time is shortened in software implementation

Engineering Contradiction:
Improveprocessing efficiencyVSAvoidprocessing time
Core Design Contradiction:
ProductivityVSLoss of time

Solution Approach 1:

The patent performs preliminary SAD calculation during the refinement processing stage, before the final BDOF decision is made. By preparing this information in advance, the system enables fast selective BDOF application without requiring additional computation during the critical rendering path, thus improving both hardware efficiency and software performance.

Inventive Principle:
Principle #10Preliminary action

Data Source

PatentEP4557733A1Image decoding device, image decoding method, and program
Publication Date: 2025.05.21 KDDI CORP
  • EP4557733A1 patent drawingFigure 1
  • EP4557733A1 patent drawingFigure 2
  • EP4557733A1 patent drawingFigure 3

AI summary

An image decoding device (200) includes: a motion vector decoding unit (241B) that decodes a motion vector from encoded data; a refinement unit (241C) that performs refinement processing to correct the decoded motion vector; and a predictive signal generation unit (241D) that generates a predictive signal on the basis of the corrected motion vector output from the refinement unit (241C), wherein the predictive signal generation unit (241D) determines whether or not to perform bidirectional optical flow, BDOF, processing for each sub-block which underwent the refinement processing, on the basis of information calculated in the course of the refinement processing, wherein the sub-block is generated by splitting a block, and wherein the predictive signal generation unit (241D) determines, for each of the sub-blocks, that the BDOF processing is not to be performed when a minimum search cost of the refinement processing is less than or equal to a predefined threshold value, wherein the search cost is a sum of absolute differences or any other indicator of the similarity between image signals, calculated in the course of the refinement processing.