Depth-Map Warping for Efficient Dependent-View Video Coding
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing multi-view video coding technologies inefficiently exploit interdependencies between views, leading to high bit rates and suboptimal motion parameter reuse across multiple camera perspectives, particularly in autostereoscopic displays.
Innovation Solution
Utilize depth maps to derive motion vector predictor candidates by determining disparity vectors between reference and dependent views, enabling motion-compensated prediction and inter-view redundancy reduction.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Productivity
If conventional multi-view video coding is used without depth maps, then the bit rate increases linearly with the number of coded views, but the coding efficiency is poor and inter-view redundancy is not sufficiently exploited
Solution Approach 1:
The depth map is generated in advance for the reference view and stored in the decoder buffer before processing the dependent view. This preliminary preparation enables the decoder to immediately utilize depth information for motion vector prediction when processing the dependent view, improving coding efficiency without requiring real-time depth computation during the encoding process.
Solution Approach 2:
The depth map acts as an intermediary data structure that bridges the reference view and dependent view. By representing the geometric relationship between views through depth information, it enables accurate motion vector prediction and disparity-compensated prediction, thereby reducing the bit rate required to transmit multiple views while maintaining high coding efficiency.
2Adaptability or versatility
If depth maps are transmitted for each view, then intermediate views can be rendered at the receiver side, but the bit rate increases due to the additional depth map data
Solution Approach 1:
The depth map of the reference view serves multiple functions: it enables motion vector prediction for dependent views, supports disparity-compensated prediction, and allows rendering of intermediate views at the receiver side. By making the depth map universally applicable across different viewing conditions, the system achieves high adaptability without proportionally increasing the bit rate.
Solution Approach 2:
Instead of transmitting depth maps for all views, the system copies and reuses the reference view's depth map through warping and transformation to generate depth information for dependent views. This copying approach eliminates the need to transmit redundant depth data for each view while still enabling accurate rendering of intermediate views.
3Productivity
If motion parameters are independently coded for each view, then decoding simplicity is maintained, but motion parameter reuse across views is suboptimal and coding efficiency is reduced
Solution Approach 1:
The system uses depth map information to generate motion vector predictions for dependent views based on the reference view's motion parameters. This feedback mechanism allows the decoder to leverage previously coded motion information and refine it using depth-based geometric relationships, improving motion parameter reuse efficiency while maintaining relatively simple decoding through index-based selection from predicted candidates.
Data Source
AI summary
The usual coding order according to which the reference view is coded prior to the dependent view, and within each view, a depth map is coded subsequent to the respective picture, may be maintained and does lead to a sacrifice of efficiency in performing inter-view redundancy removal by, for example, predicting motion data of the current picture of the dependent view from motion data of the current picture of the reference view. Rather, a depth map estimate of the current picture of the dependent view is obtained by warping the depth map of the current picture of the reference view into the dependent view, thereby enabling various methods of inter-view redundancy reduction more efficiently by bridging the gap between the views. According to another aspect, the following discovery is exploited: the overhead associated with an enlarged list of motion predictor candidates for a block of a picture of a dependent view is comparatively low compared to a gain in motion vector prediction quality resulting from an adding of a motion vector candidate which is determined from an, in disparity-compensated sense, co-located block of a reference view.