Geometry-Based Disparity Prediction for Multiview Video Coding
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Multiview data, such as multiview video and images, require significant bits for encoding and decoding, leading to increased memory storage and bandwidth requirements, making real-world applications less feasible.
Innovation Solution
Geometry-based disparity prediction determines corresponding block pairs in reconstructed images from different views, projecting them to determine disparity vector candidates and predicting disparity vectors for each coding block, reducing the number of encoded bits.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If multiview data is encoded using traditional methods, then the quality of multiview visual experience is maintained, but the number of encoded bits and memory storage requirements increase significantly
Solution Approach 1:
The patent performs preliminary disparity estimation using geometry-based methods (epipolar geometry, fundamental matrix) before actual encoding. By pre-calculating disparity vectors and identifying corresponding block pairs between views, the system prepares prediction data in advance that significantly reduces the bit rate needed for final encoding, thus resolving the contradiction between maintaining quality and reducing data quantity
Solution Approach 2:
The patent introduces geometry-based disparity vectors as an intermediary element between multiple video views. These disparity vectors serve as mediators that capture the geometric relationship between views, allowing the encoder to reference and predict content across views without transmitting all raw data, thereby reducing encoded bits while preserving visual quality
2Measurement precision
If multiview data is encoded with high fidelity, then the visual experience quality is improved, but the memory storage and bandwidth requirements become prohibitive for real-world applications
Solution Approach 1:
The patent segments the encoding process into distinct stages: geometry-based disparity estimation, corresponding block pair identification, and final encoding with references to previously decoded views. This segmentation allows each stage to operate efficiently with reduced data requirements, making the overall system more feasible for real-world applications while maintaining high visual quality
Solution Approach 2:
The patent changes the parameter representation by using geometry-based disparity vectors instead of traditional motion vectors. This parameter transformation leverages the geometric relationships between multiple views to represent data more compactly, reducing both memory storage and bandwidth requirements while preserving the fidelity needed for high-quality visual experience
Data Source
AI summary
Described herein is technology for, among other things, multiview coding with geometry-based disparity prediction. The geometry-based disparity prediction involves determining corresponding block pairs in a number of reconstructed images for an image being coded. The reconstructed images and the image represent different views of a scene at a point in time. Each corresponding block pair is projected on the image. This enables determination of disparity vector candidates. For each coding block of the image, a predicted disparity vector is determined based on the disparity vector candidates. Then, the predicted disparity vector may be utilized to obtain the bits to be encoded. The geometry-based disparity prediction reduces the number of encoded bits.