Prediction and Sampling with Local Graph Transforms for Quasi-Lossless Light Field Compression.

Area of Science:

• Computer Vision
• Image Processing
• Signal Processing

Background:

• Graph-based transforms offer powerful image energy compaction.
• High dimensionality of data like light fields poses computational challenges for large graph supports.
• Local transforms limit exploitation of long-range dependencies in spatial and angular dimensions.

Purpose of the Study:

• To develop efficient sampling and prediction schemes for local graph-based transforms.
• To enable exploitation of dependencies beyond the local graph support in high-dimensional data.
• To achieve quasi-lossless compression of light fields.

Main Methods:

• Utilized sampling and prediction schemes with local graph-based transforms.
• Focused on spatio-angular transforms for light field data.
• Investigated methods to overcome limitations of local support size.

Main Results:

• Demonstrated efficient signal energy compaction.
• Successfully exploited dependencies beyond the local graph support.
• Achieved very efficient quasi-lossless compression of light fields.

Conclusions:

• The proposed approach effectively addresses computational challenges in graph-based transforms for high-dimensional data.
• Novel schemes enable efficient energy compaction and dependency exploitation in light fields.
• The method is highly efficient for spatio-angular transforms in quasi-lossless light field compression.