Division-Free Bilateral Filtering for Video Decoding

Resolve Bottlenecks,
Find Innovative Solutions
Generate Solutions

Solution Overview

Problem

Existing bilateral filtering techniques in video encoding and decoding are computationally complex due to the number of division operations and high-precision multiplication operations required.

Innovation Solution

The proposed division-free bilateral filtering (DFBil) method reduces computational complexity by eliminating division operations and high-precision multiplications, using look-up tables and simplified weighting factors based on distance and range information.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Measurement precision

If conventional bilateral filtering is used to improve video coding quality, then filtering accuracy is improved, but computational complexity increases due to division operations and high-precision multiplications

Engineering Contradiction:
Improvefiltering accuracyVSAvoidcomputational complexity
Core Design Contradiction:
Measurement precisionVSDevice complexity

Solution Approach 1:

The patent transforms the bilateral filtering computation by changing the parameter representation: instead of using direct distance and range parameters that require division and high-precision multiplication, the invention uses transformed parameters (squared distances, scaled values) that can be computed using only integer multiplication and addition. This parameter transformation maintains filtering accuracy while eliminating complex operations.

Inventive Principle:
Principle #35Parameter changes

Solution Approach 2:

The patent substitutes the mechanical computation system by replacing division operations with multiplication operations and high-precision multiplications with integer arithmetic operations. This substitution is achieved through algebraic manipulation of the filtering formula, where the weight calculation is reformulated to avoid floating-point arithmetic while preserving the filtering effect.

Inventive Principle:
Principle #28Mechanics substitution (Replace mechanical system)

2Measurement precision

If conventional bilateral filtering is used to improve video coding quality, then filtering accuracy is improved, but processing speed decreases due to computational complexity

Engineering Contradiction:
Improvefiltering accuracyVSAvoidprocessing speed
Core Design Contradiction:
Measurement precisionVSProductivity

Solution Approach 1:

By changing the computational parameters from direct distance/range values to squared and scaled integer representations, the patent enables faster computation using integer arithmetic units that are inherently faster than floating-point division and multiplication units in video processing hardware.

Inventive Principle:
Principle #35Parameter changes

3Measurement precision

If conventional bilateral filtering is used to improve video coding quality, then filtering accuracy is improved, but energy consumption increases due to computational complexity

Engineering Contradiction:
Improvefiltering accuracyVSAvoidenergy consumption
Core Design Contradiction:
Measurement precisionVSUse of energy by moving object

Solution Approach 1:

The patent substitutes energy-intensive floating-point division and high-precision multiplication operations with low-energy integer arithmetic operations. This substitution leverages the fact that integer arithmetic requires fewer computational cycles and less power in video processing hardware, thereby reducing overall energy consumption while maintaining filtering effectiveness.

Inventive Principle:
Principle #28Mechanics substitution (Replace mechanical system)

Data Source

PatentEP3649777B1Division-free bilateral filter
Publication Date: 2025.02.12 QUALCOMM INC
  • EP3649777B1 patent drawingFigure 1
  • EP3649777B1 patent drawingFigure 2A~2b
  • EP3649777B1 patent drawingFigure 3A~3D

AI summary

A video decoder can be configured to perform a filtering operation by determining a first difference value that corresponds to a difference between the sample value of the selected sample and a sample value of a first neighboring sample; multiplying the first difference value by a first weighting parameter to determine a first weighted difference value; determining a second difference value that corresponds to a difference between the sample value of the selected sample and a sample value of a second neighboring sample; multiplying the second difference value by a second weighting parameter to determine a second weighted difference value; and adding the first weighted difference value and the second weighted difference value to the sample value of the selected sample to determine the modified sample value.