End-to-end varifocal multiview images coding framework from data acquisition end to vision application end.

Area of Science:

• Computer Vision
• Multimedia Engineering
• Image Processing

Background:

• Varifocal multiview (VFMV) data presents significant compression challenges due to inherent redundancy from dense view arrangements and focus variations.
• Existing compression methods struggle with the irregular focusing distributions and decreased inter-view similarity in VFMV data.

Purpose of the Study:

• To propose a novel end-to-end coding scheme for varifocal multiview (VFMV) images, addressing data acquisition to application-end compression.
• To enhance coding efficiency by improving inter-view similarity through a proposed view reordering technique.

Main Methods:

• VFMV data acquired via conventional imaging, plenoptic refocusing, and 3D creation.
• Implementation of a view reordering strategy based on descending irregular focusing distributions.
• Development of a 4-directional prediction (4DP) method for compressing reordered VFMV video sequences using four reference views.

Main Results:

• The proposed coding scheme significantly outperforms comparison methods in objective and subjective quality, as well as computational complexity.
• Experiments demonstrate VFMV's capability for extended depth of field compared to conventional multiview systems.
• Validation confirms the effectiveness of view reordering, superiority over MV-HEVC, and flexibility across diverse data types.

Conclusions:

• The developed end-to-end coding scheme offers a new paradigm for efficient VFMV compression.
• The proposed techniques effectively mitigate VFMV data redundancy, leading to superior compression performance.
• The VFMV compression scheme enables advanced vision applications with enhanced visual quality and extended depth of field.