Progressive Multi-Plane Images Construction for Light Field Occlusion Removal.

Area of Science:

• Computer Vision
• Image Processing
• Computer Graphics

Background:

• Light Field (LF) technology offers potential for occlusion removal by leveraging multiple views.
• Existing LF methods have limitations in implicitly modeling scenes and handling occlusions effectively.

Purpose of the Study:

• To propose a novel Progressive Multi-Plane Images (MPI) construction method for advanced Light Field-based occlusion removal.
• To improve the accuracy and effectiveness of occlusion removal by progressively modeling scenes layer by layer.

Main Methods:

• Developed a progressive MPI construction approach, building layers from near to far.
• Introduced an Occlusion-Aware Attention Network to generate MPI layers using occlusion priors from nearer layers.
• Filtered occlusions within each layer to enhance background recovery using visible views.

Main Results:

• The proposed method demonstrates superior performance over state-of-the-art LF occlusion removal techniques in quantitative and visual evaluations.
• Applied the progressive MPI construction to view synthesis, achieving significantly improved occlusion edge quality.

Conclusions:

• The novel progressive MPI construction method effectively addresses limitations in current LF occlusion removal techniques.
• The method shows promise for both occlusion removal and enhancing view synthesis quality in complex scenes.