Neural-Network Post-Processing Filters for Video Colorization

Resolve Bottlenecks,
Find Innovative Solutions
Generate Solutions

Solution Overview

Problem

Current video coding technologies, such as VVC and VSEI, do not adequately support applications requiring colorization, bit-depth increase, or frame rate changes for black and white videos, and lack clear specifications for neural-network post-filter characteristics, leading to inefficiencies and incomplete processing.

Innovation Solution

Implement specific syntax and semantics for neural-network post-processing filters (NNPF) to support 4:0:0 input, bit-depth increase, and frame rate changes, including new purposes and constraints for chroma format changes, bit-depth differences, and frame rate adjustments, ensuring complete and efficient processing.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Adaptability or versatility

If conventional video coding technologies are used, then basic video compression is achieved, but support for colorization, bit-depth increase, and frame rate changes is insufficient

Engineering Contradiction:
Improvesupport for colorization and format conversionVSAvoidprocessing completeness
Core Design Contradiction:
Adaptability or versatilityVSReliability

Solution Approach 1:

The neural-network post-processing filter is designed to perform multiple functions including colorization, chroma upsampling, and general post-processing through a unified filter structure. The bitstream contains configuration information that directs the filter to perform different operations based on the video format and processing requirements, making the system versatile while maintaining reliable processing for each specific function.

Inventive Principle:
Principle #6Universality (Multi-functionality)

Solution Approach 2:

The patent changes key parameters of the video signal including chroma format (from 4:0:0 to 4:2:0), bit depth, and frame rate through the neural-network filter. The filter is configured with specific parameters indicating the input format, output format, and desired processing type, allowing transformation between different video specifications while maintaining quality.

Inventive Principle:
Principle #35Parameter changes

2Manufacturing precision

If neural-network post-processing filter is applied for colorization, then coding quality is improved, but device complexity increases

Engineering Contradiction:
Improvecoding qualityVSAvoidfilter configuration complexity
Core Design Contradiction:
Manufacturing precisionVSDevice complexity

Solution Approach 1:

The bitstream serves as an intermediary that carries configuration information about the neural-network filter settings. Instead of complex direct control of the filter, the system uses standardized bitstream parameters to convey filter configuration, input format, output format, and processing type, simplifying the control mechanism while maintaining high coding quality.

Inventive Principle:
Principle #24Intermediary (Mediator)

Solution Approach 2:

The filter configuration information is prepared and embedded in the bitstream before video decoding. This preliminary action allows the decoder to pre-configure the neural-network filter with the correct parameters for colorization or other processing, avoiding complex runtime configuration and reducing operational complexity.

Inventive Principle:
Principle #10Preliminary action

3Adaptability or versatility

If NNPF supports multiple purposes including colorization, then functionality is diversified, but specification clarity decreases

Engineering Contradiction:
Improvefunctionality diversityVSAvoidspecification clarity
Core Design Contradiction:
Adaptability or versatilityVSDifficulty of detecting and measuring

Solution Approach 1:

The patent segments the filter functionality into distinct processing types (colorization, chroma upsampling, general post-processing) that are selectively activated based on video format and requirements. The bitstream contains specific flags and parameters that clearly indicate which processing type to use, making the specification clear despite the diversity of supported functions.

Inventive Principle:
Principle #1Segmentation

Solution Approach 2:

The neural-network filter is designed to dynamically adapt its operation based on configuration parameters in the bitstream. The same filter infrastructure can perform different functions (colorization, upsampling, etc.) by changing its operational mode based on input parameters, providing versatility while maintaining a clear, unified specification structure.

Inventive Principle:
Principle #15Dynamics

Data Source

PatentUS20250337961A1Method, apparatus, and medium for video processing
Publication Date: 2025.10.30 BYTEDANCE INC
  • US20250337961A1 patent drawing
  • US20250337961A1 patent drawing
  • US20250337961A1 patent drawing

AI summary

Embodiments of the present disclosure provide a solution for video processing. A method for video processing is proposed. The method comprises: performing a conversion between a video and a bitstream of the video, wherein a neural-network post-processing filter (NNPF) is applied on at least one picture associated with the video, the bitstream comprises a first indication indicating a purpose of the NNPF, and one of candidates for the purpose is colorization of the at least one picture.