Reference View Selection in DIBR-Based Multiview Coding.

Area of Science:

• Computer Vision
• Multimedia Systems
• Data Compression

Background:

• Emerging applications like augmented reality and 3D scenes demand large image datasets, increasing data volume and resource requirements.
• Multiview video systems face challenges in data representation and transmission for high-quality experience in resource-constrained environments.
• Current compression strategies often rely on reference views but lack optimal selection methods for number and placement.

Purpose of the Study:

• To address the overlooked problem of optimal reference view selection in multiview coding.
• To develop an algorithm for selecting the ideal number and positions of reference views.
• To improve rate-distortion performance in multiview video compression.

Main Methods:

• Proposed a novel metric to quantify similarity between views.
• Formulated an optimization problem to minimize reconstruction distortion and coding rate.
• Solved the optimization problem using a shortest path algorithm.
• Experimentally validated the solution in distributed coding and 3D-HEVC systems.

Main Results:

• The proposed algorithm effectively determines the optimal number and placement of reference views.
• Considering 3D scene geometry in reference view selection yields significant rate-distortion improvements.
• The new strategy outperforms traditional methods based solely on camera distance.

Conclusions:

• Optimal reference view selection is critical for efficient multiview video compression.
• The proposed shortest path algorithm provides a robust solution for reference view positioning.
• This approach enhances quality of experience in resource-limited multimedia applications.