Image Encoding Reference Picture Segmentation for Random Access
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Conventional image encoding devices face a trade-off between coding efficiency and random access intervals, where shortening random access intervals decreases the percentage of inter pictures usable for motion-compensated prediction, and increasing inter pictures to improve efficiency lengthens these intervals, making it difficult to enhance prediction efficiency while maintaining short random access intervals.
Innovation Solution
An image encoding device that partitions input images into blocks, determines a coding mode for each block, and performs encoding using a prediction image generator to set a reference picture from nearby intra pictures for motion-compensated prediction, encoding identification information to enable random access and improve coding efficiency.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Ease of operation
If the random access intervals are shortened to enable random access from inter pictures, then the ease of operation is improved, but the coding efficiency deteriorates due to decreased percentage of usable inter pictures for motion-compensated prediction
Solution Approach 1:
The patent segments the reference picture list into two distinct parts: long-term reference pictures (LTRP) and short-term reference pictures (STRP). LTRPs serve as stable anchor points for random access, while STRPs provide high-correlation references for motion-compensated prediction. This segmentation allows the system to simultaneously maintain short random access intervals through LTRPs and preserve coding efficiency through STRPs, resolving the contradiction between ease of operation and productivity.
2Productivity
If inter pictures are increased to improve coding efficiency, then the productivity is improved, but the random access intervals are lengthened making random access more difficult
Solution Approach 1:
By dividing reference pictures into long-term and short-term categories, the patent enables the system to use LTRPs for random access points (maintaining short intervals) while using STRPs for high-efficiency prediction (improving coding efficiency). This segmentation resolves the contradiction by allowing both types of pictures to coexist with distinct functional roles.
Solution Approach 2:
The patent dynamically manages the reference picture list by allowing LTRPs to be updated periodically while STRPs are updated more frequently based on prediction needs. This dynamic management enables the system to adaptively balance between maintaining short random access intervals and preserving coding efficiency, depending on the specific encoding context.
3Manufacturing precision
If motion-compensated prediction uses multiple reference pictures to improve prediction accuracy, then the manufacturing precision is improved, but the device complexity increases due to dependence in decoding
Solution Approach 1:
The patent segments the prediction process into two stages: first using LTRPs for base prediction to ensure decodability, then optionally using STRPs for refinement to improve accuracy. This segmented approach maintains prediction accuracy while reducing decoding complexity by establishing a clear hierarchical dependency structure where LTRP-based prediction forms the independent foundation.
Solution Approach 2:
The patent applies different reference picture qualities to different regions or blocks within the current picture. Certain blocks may use LTRP-based prediction for regions requiring random access compatibility, while other blocks use STRP-based prediction for regions where higher accuracy is prioritized. This local differentiation resolves the contradiction by allowing high prediction accuracy where needed without creating system-wide decoding dependence.
Data Source
AI summary
When a randomly-accessible inter picture is encoded, a prediction image generator sets a reference picture from among a plurality of randomly-accessible intra pictures, and performs a motion-compensated prediction using the set reference picture for a prediction process, and a variable length encoding unit 13 encodes both picture position information showing the position of the reference picture, and identification information showing that the randomly-accessible inter picture is randomly accessible, and multiplexes encoded data about both the picture position information and the identification information into a bitstream.