Reference Image List Switching for Low-Bitrate Image Decoding
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing image encoding and decoding technologies face challenges in achieving efficient compression at low bit rates, particularly in handling reference image lists and motion estimation for accurate prediction.
Innovation Solution
The proposed method involves modifying reference image lists by adding or removing specific types of reference images and adjusting the order of reference images to address the technical problem, and the use of hierarchical structures to address the technical problem, and the use of hierarchical structures to address the technical problem, by modifying reference image lists and using a hierarchical structure for image encoding and decoding.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If motion estimation encoding is performed using reference images with different types (e.g., long-term and short-term references), then prediction accuracy is improved, but device complexity and processing overhead increase due to managing multiple reference image lists
Solution Approach 1:
The patent divides the reference image list into multiple types (first type and second type of reference images) with different retention policies. Short-term references are kept in the first type for immediate prediction, while long-term references are managed in the second type. This segmentation allows the system to use both types for accurate prediction without treating all references uniformly, thus improving prediction accuracy while organizing complexity through structured categorization.
Solution Approach 2:
The patent implements dynamic management of reference image lists where the composition and ordering of references change based on prediction needs and block group indicators. The system dynamically selects and modifies reference image lists for different block groups, allowing flexibility in using appropriate reference types for different prediction scenarios, thereby maintaining accuracy while adapting to varying complexity requirements.
2Productivity
If multiple reference image lists are maintained and modified for different block groups, then compression efficiency at low bit rates is improved, but processing time and computational load increase
Solution Approach 1:
The patent performs preliminary organization of reference images into typed lists during the encoding process. By pre-categorizing reference images into first and second types with different retention characteristics, the system prepares the reference structure in advance, reducing the computational burden during actual prediction operations and enabling efficient low-bit-rate compression without excessive processing time during real-time encoding.
Solution Approach 2:
The patent changes the parameters of reference image management by introducing type-based classification and differential retention policies. Instead of uniformly managing all reference images, the system applies different parameters (retention duration, list modification rules) to different reference types, enabling optimized compression efficiency at low bit rates while controlling processing time through parameterized management strategies.
3Measurement precision
If reference images are retained longer in memory for future predictions, then prediction accuracy for future blocks is improved, but memory usage and data redundancy increase
Solution Approach 1:
The patent segments the reference image storage into two distinct categories: first type reference images with shorter retention and second type reference images with longer retention. This segmentation allows the system to keep essential recent references readily available while maintaining older long-term references separately, improving prediction accuracy for future blocks without uniformly increasing memory usage for all reference types.
Solution Approach 2:
The patent implements a selective discarding and recovering mechanism where first type reference images are discarded after short-term use, while second type reference images are retained for long-term recovery and reuse. This approach allows the system to manage memory efficiently by discarding short-term references that are no longer needed while recovering and reusing long-term references when beneficial, thus improving future prediction accuracy without proportionally increasing overall memory usage.
Data Source
AI summary
An image decoding method may include: obtaining, from a sequence parameter set of a bitstream, information indicating a plurality of first reference image lists for an image sequence including a current image; obtaining, from a group header of the bitstream, an indicator for a current block group including a current block in the current image; obtaining a second reference image list by modifying a first reference image list indicated by the indicator, from among the plurality of first reference image lists, into the second reference image list; and prediction-decoding a lower block of the current block based on a second reference image included in the second reference image list.