Arbitrarily shaped motion prediction for depth video compression using arithmetic edge coding.

Area of Science:

• Computer Vision
• Image Processing
• Video Compression

Background:

• Depth image compression is crucial for 3D visual data representation in texture-plus-depth formats.
• Depth information aids applications like object identification and segmentation.
• Existing methods face challenges in efficiently encoding depth maps with distinct foreground and background regions.

Purpose of the Study:

• To develop an efficient depth video compression technique leveraging motion characteristics of similar depth pixels.
• To minimize the overhead associated with arbitrarily shaped sub-block motion prediction.
• To improve the overall bitrate reduction for depth image compression.

Main Methods:

• Dividing depth blocks into arbitrarily shaped sub-blocks along boundaries for separate motion prediction.
• Developing Arithmetic Edge Coding (AEC) to efficiently encode these boundaries by incorporating geometric correlation.
• Implementing two optimization procedures: lossy boundary compression within code blocks and rate-distortion optimized trellis across code blocks.

Main Results:

• Achieved an average overall bitrate reduction of up to 33% compared to classical H.264/AVC.
• Demonstrated efficient encoding of arbitrarily shaped sub-block prediction residuals.
• Successfully minimized the overhead for transmitting dividing boundaries.

Conclusions:

• The proposed method significantly enhances depth video compression efficiency.
• Arithmetic Edge Coding with optimization procedures effectively reduces bitrate for depth map representation.
• This technique offers a promising solution for compact 3D visual data transmission and processing.