Highly Efficient Multiview Depth Coding Based on Histogram Projection and Allowable Depth Distortion.

Area of Science:

• Computer Vision
• Video Compression
• 3D Graphics

Background:

• Mismatches in 3D video systems lead to redundant depth map representations.
• Existing methods struggle with precision inconsistencies in disparity, depth value, and rendering position.

Purpose of the Study:

• To propose a highly efficient multiview depth coding scheme.
• To reduce redundancies and improve compression performance in depth map representations.
• To maintain lossless view synthesis quality.

Main Methods:

• Developed a Depth Histogram Projection (DHP) technique to exploit sparse depth map representations.
• Introduced an Allowable Depth Distortion (ADD) model to address depth value and rendering position mismatches.
• Provided a mathematical foundation for DHP-based lossless depth coding through rate-distortion cost analysis.

Main Results:

• DHP-based depth coding achieved an average bit rate saving of 20.66% to 19.52% for lossless coding on MV-HEVC.
• DHP and ADD combined achieved average depth bit rate reductions of 46.69%, 34.12%, and 28.68% for integer, half, and quarter pixel precisions, respectively.
• Similar gains were observed on 3D-HEVC, HEVC Intra coding, and JPEG2000 platforms.

Conclusions:

• The proposed DHP and ADD scheme offers significant compression efficiency for depth coding in 3D video.
• The method effectively maintains lossless view synthesis quality across various rendering precisions.
• This approach provides a robust solution for optimizing depth map representations in multiview video coding.