Efficient multiview depth coding optimization based on allowable depth distortion in view synthesis.

Area of Science:

• Computer Vision
• Image Processing
• 3D Graphics

Background:

• Depth video is crucial for 3D view synthesis, providing geometrical scene information.
• A mismatch in depth and disparity levels can cause rendering position errors in synthesized views.
• Depth distortion and rendering errors exhibit a many-to-one mapping, where different distortions yield the same error.

Purpose of the Study:

• To develop an allowable depth distortion (ADD) model for efficient 3D depth map coding.
• To minimize view synthesis distortion under bit rate constraints using an ADD-based rate-distortion model.
• To reduce depth bit rates without compromising synthesized image quality.

Main Methods:

• Modeling the relationship between depth distortion and rendering position error as a many-to-one function.
• Proposing an allowable depth distortion (ADD) model for 3D depth map coding.
• Developing an ADD-based rate-distortion model for mode decision and motion/disparity estimation.
• Implementing an ADD-based depth bit reduction algorithm.

Main Results:

• Achieved Bjontegaard delta peak signal-to-noise ratio gains of 1.58 dB (half-pixel) and 2.68 dB (integer-pixel) in intra depth coding.
• Demonstrated high efficiency for inter depth coding across various metrics.
• Successfully reduced depth bit rates while maintaining synthesized image quality.

Conclusions:

• The proposed ADD model and associated algorithms offer significant improvements in 3D depth map coding efficiency.
• The method effectively balances bit rate reduction with the quality of synthesized 3D views.
• The approach is beneficial for both intra and inter depth coding scenarios.