Gradual Decoding Refresh with CTU-Aligned Picture Boundaries
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing video coding systems face challenges in maintaining content accuracy and efficiency during decoding refresh, particularly in aligning coding tree units with refreshed and unrefreshed areas, leading to potential errors and inefficiencies.
Innovation Solution
The proposed solution involves defining a refresh period for coded pictures, aligning the boundary between refreshed and unrefreshed areas with coding tree units, and disabling prediction processes across these boundaries to ensure accurate decoding from a random access point, with specific encoding techniques for both areas.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Adaptability or versatility
If the boundary between refreshed and unrefreshed areas is not aligned with coding tree unit boundary, then the decoding flexibility is improved, but decoding accuracy deteriorates due to prediction errors across boundaries
Solution Approach 1:
The picture is segmented into refreshed and unrefreshed areas with boundaries aligned to coding tree unit boundaries. This segmentation ensures that prediction processes are constrained within each segment, preventing prediction errors from propagating across segment boundaries while maintaining decoding flexibility within each segment.
Solution Approach 2:
Different decoding constraints are applied to different areas of the picture based on their refresh status. Refreshed areas use standard prediction processes, while unrefreshed areas have prediction processes disabled or constrained. This local differentiation maintains accuracy in refreshed areas while preserving flexibility in unrefreshed areas.
2Productivity
If prediction processes are enabled across coding tree unit boundaries, then the encoding efficiency is improved, but content accuracy deteriorates in refreshed areas
Solution Approach 1:
The coding picture is divided into refreshed and unrefreshed regions with boundaries aligned to coding tree unit boundaries. Prediction processes are constrained to operate only within refreshed areas or from refreshed to unrefreshed areas, preventing inaccurate predictions from propagating while maintaining encoding efficiency through selective prediction application.
Solution Approach 2:
Prediction processes are enabled with different constraints in different regions. In refreshed areas, full prediction is allowed for high accuracy. In unrefreshed areas, prediction is constrained or disabled to prevent error propagation. This local differentiation resolves the contradiction between efficiency and accuracy.
3Loss of energy
If the refresh period is extended, then the bitrate efficiency is improved, but the decoding complexity increases due to larger refreshed area management
Solution Approach 1:
The picture is segmented into refreshed and unrefreshed areas with boundaries aligned to coding tree unit boundaries, enabling independent processing of each segment. This segmentation simplifies decoding complexity by allowing parallel or sequential processing of manageable segments while maintaining extended refresh periods for bitrate efficiency.
Solution Approach 2:
The boundary between refreshed and unrefreshed areas is dynamically adjusted based on the refresh period and picture content. This dynamic adjustment allows the system to optimize the balance between refresh coverage (affecting bitrate efficiency) and processing complexity, adapting to different encoding scenarios.
Data Source
AI summary
The present embodiments relate to a method for encoding and a technical equipment for implementing the method. The method includes receiving a sequence of pictures to be encoded; defining a refresh period for coded pictures, wherein the decoded recovery point picture and all subsequent decoded pictures in output order are correct in content when starting decoding from the random access point picture. The encoding further includes determining the size of the refreshed area; selecting a number of sample columns or rows outside a conformance cropping window so that a boundary between the refreshed area and the unrefreshed area of the picture is aligned with a coding tree unit boundary.