Video Predictive Decoding Using Picture Type Classification

Resolve Bottlenecks,
Find Innovative Solutions
Generate Solutions

Solution Overview

Problem

Conventional video predictive decoding technologies discard all leading pictures as incorrectly decoded, even though some can be correctly decoded, leading to suboptimal prediction performance due to the lack of a method to identify decodable pictures for use as reference.

Innovation Solution

A system that classifies pictures into three types: CRA pictures, type 1 pictures, and type 2 pictures, allowing the identification and utilization of correctly decodable type 2 pictures as reference for subsequent decoding, thereby improving prediction performance by selectively discarding only non-decodable pictures.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Reliability

If all leading pictures are discarded as incorrectly decoded, then decoding reliability is improved, but prediction performance deteriorates due to loss of potentially useful reference pictures

Engineering Contradiction:
Improvedecoding reliabilityVSAvoidprediction performance
Core Design Contradiction:
ReliabilityVSManufacturing precision

Solution Approach 1:

The patent segments leading pictures into two categories: decodable and non-decodable. By introducing picture type flags and classification mechanisms, the system divides the previously uniform treatment of all leading pictures into differentiated handling, allowing decodable pictures to be retained as reference while discarding only non-decodable ones.

Inventive Principle:
Principle #1Segmentation

Solution Approach 2:

The patent implements feedback mechanisms through picture type flags and classification information that provide decoding status back to the reference picture selection process. This feedback loop enables the system to adjust reference picture selection based on actual decoding outcomes, improving both reliability and prediction performance.

Inventive Principle:
Principle #23Feedback

2Reliability

If all leading pictures are discarded, then errors from incorrectly decoded pictures are prevented, but compression efficiency deteriorates due to reduced reference picture availability

Engineering Contradiction:
Improveerror preventionVSAvoidcompression efficiency
Core Design Contradiction:
ReliabilityVSProductivity

Solution Approach 1:

The patent segments leading pictures into decodable and non-decodable categories, enabling selective retention. This segmentation allows the system to maintain error prevention by discarding non-decodable pictures while preserving compression efficiency by utilizing decodable pictures as reference pictures for inter-frame prediction.

Inventive Principle:
Principle #1Segmentation

Solution Approach 2:

The patent changes the parameter of picture utilization by introducing picture type classification and flags. This parameter change transforms the binary discard/retain decision into a nuanced selection process based on decodability status, thereby improving compression efficiency without compromising error prevention.

Inventive Principle:
Principle #35Parameter changes

3Manufacturing precision

If picture type classification is implemented, then prediction performance is improved by selective reference picture usage, but device complexity increases

Engineering Contradiction:
Improveprediction performanceVSAvoidsystem complexity
Core Design Contradiction:
Manufacturing precisionVSDevice complexity

Solution Approach 1:

The patent segments the picture processing system into classified categories with associated flags and types. While this segmentation improves prediction performance by enabling selective reference picture usage, it inherently increases device complexity through the introduction of classification logic, flag management, and type checking mechanisms.

Inventive Principle:
Principle #1Segmentation

Solution Approach 2:

The patent implements self-service mechanisms where the encoding device automatically classifies pictures and sets appropriate flags during the encoding process. This self-service approach reduces the need for complex external control systems, thereby mitigating some of the complexity increase while maintaining improved prediction performance.

Inventive Principle:
Principle #25Self-service

Data Source

PatentEP4221216A1Dynamic image predictive decoding method
Publication Date: 2023.08.02 NTT DOCOMO INC
  • EP4221216A1 patent drawingFigure 1
  • EP4221216A1 patent drawingFigure 2
  • EP4221216A1 patent drawingFigure 3

AI summary

A video predictive decoding method is provided comprising: inputting a bitstream including compressed picture data of a plurality of pictures constituting a video sequence, where each picture has a network abstraction layer unit type that identifies said picture as one of a plurality of picture types including a clean random access picture (CRA), a random access skipped (RAS) picture and a non-RAS picture; decoding the compressed picture data to reconstruct the plurality of pictures based on their picture types; and outputting the reconstructed pictures.