An Occlusion Compensation Learning Framework for Improving the Rendering Quality of Light Field

Area of Science:

• Computer Vision
• Computer Graphics
• Machine Learning

Background:

• Occlusions in light field rendering (LFR) lead to incomplete or inaccurate 3D spatial structures.
• Existing LFR methods often overlook occlusions from objects not directly involved in rendering.

Purpose of the Study:

• To propose an occlusion probability learning framework (OPLF) for improving LFR quality.
• To compensate for occluded information in LFR using a novel deep learning approach.

Main Methods:

• Developed an occlusion probability learning framework (OPLF) utilizing a deep Boltzmann machine (DBM).
• Modeled occlusion probability density and visibility scores as hidden variables.
• Incorporated camera configuration and object relationships into the occlusion probability model.

Main Results:

• The OPLF effectively learns occlusion relationships between cameras and objects across multiple layers.
• Experimental comparisons demonstrate that the OPLF outperforms existing occlusion quantization schemes.
• The proposed framework significantly optimizes light field rendering quality.

Conclusions:

• The OPLF provides a robust solution for handling occlusions in light field rendering.
• This framework advances the state-of-the-art in light field reconstruction and occlusion handling.
• The proposed method offers a promising direction for future research in 3D scene reconstruction and rendering.