Local low rank approximation with a parametric disparity model for light field compression.

Area of Science:

• Computer Vision
• Image Processing
• Data Compression

Background:

• Light field data presents significant dimensionality challenges for compression.
• Existing methods struggle with efficient dimensionality reduction and accurate disparity estimation.
• Low-rank approximation is a promising technique for exploiting redundancies in light field data.

Purpose of the Study:

• To develop an efficient light field compression method.
• To introduce a parametric disparity model for improved super-pixel alignment.
• To leverage local low-rank approximation for dimensionality reduction.

Main Methods:

• A local low-rank approximation method utilizing a parametric disparity model.
• Super-ray construction defining local support for approximation, ensuring super-pixel shape/size consistency.
• Parametric disparity model estimation, including a novel low-rank prior method.
• Singular Value Decomposition (SVD) for low-rank matrix approximation on disparity-compensated super-rays.

Main Results:

• Achieved significant rate savings: up to 92.61% for real light fields (Lytro Illum) and 37.72% for synthetic data.
• Demonstrated superior performance compared to JPEG Pleno and HEVC-based light field compression schemes.
• The proposed parametric disparity model improved super-pixel alignment, enhancing low-rank approximation effectiveness.

Conclusions:

• The proposed method offers substantial improvements in light field compression efficiency.
• Parametric disparity modeling is key to effective low-rank approximation in light fields.
• This approach provides a viable solution for compressing complex light field data.