Multi-View Video Signal Processing Using Depth-Based Block Partitioning
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current video compression techniques face challenges in efficiently encoding and decoding multi-view video signals, particularly in deriving disparity vectors and performing inter-view prediction, which affects the compression efficiency of high-resolution and ultra-high resolution image data.
Innovation Solution
The method involves determining prediction and partition modes for depth blocks, obtaining motion vectors based on these modes, and performing inter prediction using depth-based block partitioning to generate prediction blocks, which are then combined according to a partition pattern derived from depth values.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Productivity
If conventional video compression techniques are used for multi-view video signals, then basic compression is achieved, but inter-view prediction efficiency and compression ratio are insufficient
Solution Approach 1:
The current block is divided into multiple sub-blocks for independent motion compensation processing. Each sub-block can have its own motion vector, allowing fine-grained prediction that adapts to local motion variations within the block, thereby improving prediction accuracy without requiring overly complex global models.
Solution Approach 2:
The patent introduces inter-view prediction by utilizing depth information as an additional dimension. Depth values from depth maps are used to calculate disparity vectors that enable prediction between different views, transforming a 2D spatial prediction problem into a 3D spatiospatial problem that exploits the third dimension (depth/parallax) for enhanced compression.
2Manufacturing precision
If high-resolution and ultra-high resolution image data is transmitted or stored using conventional methods, then image quality is maintained, but transmission and storage costs increase significantly
Solution Approach 1:
Motion vectors and disparity vectors are derived and prepared in advance during the encoding process. These vectors are stored as side information and used during decoding to reconstruct high-resolution frames from reference frames, avoiding the need to transmit all high-resolution data explicitly and reducing transmission costs.
Solution Approach 2:
Depth maps serve as intermediary data that facilitate inter-view prediction. Instead of transmitting redundant multi-view high-resolution content, depth information is used as a mediator to generate synthetic views through view synthesis, significantly reducing the amount of data that needs to be transmitted or stored.
3Productivity
If inter-view prediction is performed using conventional techniques, then some compression is achieved, but disparity vector derivation efficiency is insufficient
Solution Approach 1:
The system uses its own depth information and motion vectors to automatically derive disparity vectors without requiring external assistance or complex iterative optimization. The disparity vector derivation leverages already-available depth maps and motion compensation results, making the process self-sufficient and computationally efficient.
Solution Approach 2:
The patent changes the parameter representation by using depth values directly to compute disparity vectors through a straightforward mathematical relationship. Instead of performing complex feature matching or optimization to find correspondences between views, the system transforms depth parameters into disparity parameters using a direct formula, significantly reducing computation time.
Data Source
AI summary
The present invention provides a method and apparatus for processing a multi-view video signal, the method including: determining a prediction mode of a current depth block; determining a partition mode of the current texture block when the current texture block is a block encoded in an inter mode; obtaining a motion vector of each partition based on the partition mode; and performing inter prediction on the current texture block by using the motion vector through depth-based.