Advanced Scalability for Light Field Image Coding.

Area of Science:

• Computer Vision
• Image Processing
• Data Compression

Background:

• Light field imaging captures spatial and angular data, enhancing immersion but generating large, redundant datasets.
• Current compression methods prioritize efficiency, often neglecting crucial features like scalability.
• Scalability in light field compression is vital for adapting to various display technologies and network conditions.

Purpose of the Study:

• To propose a novel light field image compression method with high compression efficiency.
• To enable multiple scalability features: viewport, quality, and spatial.
• To ensure random access and uniform quality distribution across viewports.

Main Methods:

• Dividing light fields into sequential viewport layers for each spatial resolution.
• Encoding viewports using previously encoded ones, synthesizing intermediate views with a deep learning network.
• Employing image super-resolution to enhance lower spatial resolution layers and using them as references for higher layers.

Main Results:

• The proposed method achieves high compression efficiency.
• It successfully enables viewport, quality, and spatial scalability.
• The method demonstrates adaptability to various display types, network conditions, and user requirements.

Conclusions:

• The novel compression method offers significant improvements in light field data handling.
• It provides enhanced flexibility for light field streaming, random access, and error resilience.
• The approach effectively balances compression efficiency with essential scalability features for diverse applications.