On dependent bit allocation for multiview image coding with depth-image-based rendering.

Area of Science:

• Computer Vision
• Image Processing
• 3-D Graphics

Background:

• Multiview imaging and depth-image-based rendering (DIBR) are crucial for 3-D visual communication.
• Efficient encoding of texture and depth maps is essential for minimizing visual distortion in synthesized views.

Purpose of the Study:

• To develop an efficient bit allocation strategy for texture and depth maps in multiview images.
• To minimize visual distortion in constructed intermediate views by optimizing view selection and bit distribution.

Main Methods:

• Derived a cubic distortion model for low-complexity visual quality estimation of synthesized views.
• Formulated the optimal view selection and quantization as a shortest path problem on a 3-D trellis.
• Implemented pruning strategies based on monotonicity assumptions to reduce computational complexity.

Main Results:

• The proposed method achieves up to 1.5 dB improvement over schemes with constant quantization levels.
• The computational complexity is reduced by at least 80% compared to a full search solution.
• Demonstrated superior performance in selecting coded views and quantizing texture/depth maps.

Conclusions:

• The developed method provides an efficient and effective approach for bit allocation in multiview image encoding for DIBR.
• Optimizing view selection and quantization significantly enhances visual quality and reduces system complexity.
• The cubic distortion model and trellis-based optimization offer a practical solution for 3-D visual communication systems.