Image Encoding Apparatus Memory Optimization via Virtual Reference Extension
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Conventional image coding methods for high-resolution video require large memory and high memory access loads, limiting the ability to find optimal motion vectors, especially when memory access is restricted, leading to inefficient coding and decoding processes.
Innovation Solution
An image coding device that divides each picture into coding blocks for motion-compensated prediction and uses a motion compensation unit to generate prediction images, with a variable length coding unit that includes a reference image restriction flag to extend the reference image area virtually when necessary, allowing for efficient motion vector search and coding even with limited memory.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Quantity of substance
If the reference image memory area is limited for each motion search process unit, then the memory size and memory access load are reduced, but the ability to search for optimal motion vectors deteriorates
Solution Approach 1:
The picture is divided into multiple tiles, and each tile is further divided into coding blocks that are processed in parallel. Each motion search process unit operates on a specific tile with a limited reference image memory area, enabling parallel processing while maintaining manageable memory requirements for each unit.
Solution Approach 2:
The patent introduces a virtual extension dimension by allowing motion search process units to access reference image areas outside their designated tile boundaries. This virtual reference image area extension enables optimal motion vector search without requiring physical memory expansion for each unit.
2Measurement precision
If the reference image memory area is extended for each motion search process unit, then the motion vector search accuracy is improved, but the memory size and memory access load increase
Solution Approach 1:
By segmenting the reference image access into tile-based units with shared reference memory, the system achieves accurate motion vector search without requiring each unit to have a dedicated large memory area. The reference image is shared across multiple motion search process units.
Solution Approach 2:
The reference image memory serves multiple motion search process units simultaneously, enabling each unit to access the reference image for accurate motion vector search while sharing the same physical memory resource, thus avoiding memory duplication.
3Measurement precision
If coding is performed on high resolution video with conventional methods, then the coding accuracy is improved, but the memory size and processing load increase significantly
Solution Approach 1:
The high resolution picture is divided into multiple tiles and coding blocks that can be processed in parallel. This segmentation reduces the processing load on each individual unit while maintaining the overall coding accuracy through coordinated processing of all blocks.
Solution Approach 2:
The system dynamically adjusts the reference image area access based on the coding block being processed. When a coding block requires reference data outside its tile boundary, the system dynamically extends the virtual reference image area, optimizing memory access patterns for high resolution video coding.
Data Source
Figure 1(A)~1(C)
Figure 2
Figure 3
AI summary
Disclosed is an image decoding method including a variable length decoding step of variable-length-decoding coded data multiplexed into a bitstream to acquire compressed data associated with a coding block, a reference image restriction flag indicating whether or not to restrict a significant reference image area which is an area on a reference image which can be used for the above-mentioned motion-compensated prediction to a predetermined area, and motion information about a motion vector, a motion-compensated prediction step of carrying out a motion-compensated prediction process on the above-mentioned coding block on the basis of the above-mentioned motion information to generate a prediction image, a difference image generating step of generating a pre-compressed difference image from the compressed data associated with the above-mentioned coding block, and a decoded image generating step of adding the above-mentioned difference image and the above-mentioned prediction image to generate a decoded image, in which the above-mentioned motion-compensated prediction step includes a step of, when generating the above-mentioned prediction image, and the abode-mentioned prediction image includes a pixel located outside the significant reference image area, carrying out a predetermined extending process to generate the prediction image on the basis of the above-mentioned reference image restriction flag by using the above-mentioned motion information.