Reference List Generation for Multi-Viewpoint Video Coding Efficiency
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
In multi-viewpoint coding, existing technologies face challenges in improving coding efficiency due to difficulties in arranging reference images in the view direction within the temporal list, leading to decreased performance even when the insertion position of reference images does not need to be changed, and requiring unnecessary transmission of slice headers.
Innovation Solution
An image processing device and method that generates a reference list by inserting reference images in a layer direction between time-direction reference images, allowing for flexible arrangement of reference images in the temporal list, including options like alternating orders and specific insertion positions, to enhance coding efficiency.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Device complexity
If reference images in the view direction are arranged after reference images in the time direction in the temporal list, then the temporal list structure is simple, but it becomes difficult to raise the preferential order of view-direction reference images and refer to them effectively
Solution Approach 1:
The patent segments the reference image arrangement into two distinct lists: a temporal list maintaining simple time-direction ordering, and a reference list that selectively incorporates view-direction reference images at optimized positions. This segmentation allows each list to serve its specific purpose without compromising the other.
Solution Approach 2:
The patent introduces an intermediary mechanism (the reference list) that mediates between the simple temporal list structure and the need for efficient view-direction reference image access. The reference list acts as a bridge, selecting and repositioning relevant reference images without disrupting the temporal list's simplicity.
2Ease of operation
If slice header transmission is used to designate insertion position of view-direction reference images, then reference images can be inserted at preferable positions, but coding efficiency decreases due to unnecessary transmission overhead
Solution Approach 1:
The patent extracts the reference image insertion logic from the slice header transmission mechanism. Instead of transmitting slice headers to designate insertion positions, the patent generates the reference list through automated selection and arrangement processes, removing the unnecessary transmission overhead while preserving insertion flexibility.
Solution Approach 2:
The system performs self-service by automatically generating and managing the reference list without requiring external slice header commands. The decoding device autonomously identifies view-direction reference images and inserts them at optimal positions in the reference list, eliminating the need for transmission-based control.
3Ease of manufacture
If reference images are arranged strictly in time direction order, then the temporal list generation is straightforward, but view-direction correlation cannot be effectively utilized
Solution Approach 1:
The patent adds another dimension (view direction) to the reference image arrangement by creating a reference list that incorporates both time-direction and view-direction reference images. This multi-dimensional arrangement allows the system to maintain straightforward temporal list generation while simultaneously utilizing view-direction correlation for improved prediction accuracy.
Data Source
AI summary
An image processing device and an image processing method capable of improving the coding efficiency in multi-viewpoint coding, the image processing device being configured to generate a temporal list (RefPicListTemp0[rIdx]) of L0 in order of short-term (before) reference images of indexes 0 and 1, inter-view reference images of indexes 0 to 3, short-term (after) reference images of indexes 0 and 1, and a long-term reference image of index 0. In such a case, according to num_ref_idx_l0_active_minus1=4, a reference list of L0 is generated in order of the short-term (before) reference image of index 0 and the inter-view reference images of indexes 0 to 3.