R Picture Type for Random Access Video Encoding Efficiency
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
The H.264/AVC standard's use of IDR pictures for random access degrades encoding efficiency by limiting bidirectional motion estimation for B pictures before IDR and restricting P pictures to single IDR as reference, leading to reduced encoding efficiency for these frames.
Innovation Solution
Introducing a new picture type, R picture, which supports random access and allows inter-prediction using multiple reference pictures, including past and future frames, to enhance encoding efficiency and enable bidirectional motion estimation.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If IDR picture is used for random access, then random decoding is enabled, but bidirectional motion estimation cannot be performed for B pictures before IDR
Solution Approach 1:
The patent introduces a new picture type (RA picture) that dynamically adjusts reference picture management to enable bidirectional motion estimation for B pictures while maintaining random access capability. The system dynamically selects between unidirectional and bidirectional prediction modes based on the presence of future reference pictures in the DPB, thereby resolving the contradiction between random access reliability and encoding efficiency.
2Reliability
If DPB is cleared before IDR picture, then random access is supported, but P pictures after IDR can use only one IDR picture as reference
Solution Approach 1:
The patent implements dynamic reference picture list management where the RA picture type enables P pictures to adaptively use multiple reference pictures from the DPB. The system dynamically constructs reference picture lists based on picture timing and DPB content, allowing P pictures to select optimal reference pictures for motion estimation, thereby improving encoding efficiency while preserving random access functionality.
3Productivity
If multiple reference pictures are used for inter-prediction, then encoding efficiency is improved, but memory management complexity increases
Solution Approach 1:
The patent changes the parameter of picture type designation to introduce a new RA picture type that fundamentally alters reference picture management behavior. This parameter change enables the system to distinguish between traditional IDR pictures and new RA pictures, allowing differentiated handling of reference picture lists and memory management strategies, thereby achieving improved encoding efficiency without proportionally increasing complexity.
4Reliability
If unidirectional motion estimation is used for B pictures before IDR, then random access is maintained, but encoding efficiency is reduced
Solution Approach 1:
The patent implements dynamic prediction mode selection where B pictures can switch between unidirectional and bidirectional motion estimation based on the availability of future reference pictures. The system dynamically evaluates the DPB state and selects the optimal prediction mode for each B picture, enabling bidirectional estimation when beneficial while maintaining random access capability through the RA picture type.
Data Source
AI summary
A video encoding method using inter-prediction, includes: encoding a first picture, wherein the first picture serves as a random access picture and is encoded without referring to a reference picture which has already encoded and then decoded; inter-prediction encoding a block included in a second picture, which is encoded after the first picture and displayed before the first picture, by using a plurality of reference pictures; generating reference picture information representing one or more past pictures which are encoded before the first picture and are used as reference pictures for encoding the second picture; and transmitting the encoded first picture and the encoded second pictures and the reference picture information, wherein, when the one or more past pictures are used as the reference pictures for encoding the second picture, the one or more past pictures are deleted from a memory after encoding at least the second picture.