A Maximum Likelihood Approach for Depth Field Estimation Based on Epipolar Plane Images.

Area of Science:

• Computer Vision
• Image Processing
• Computational Photography

Background:

• Consumer electronics advancements have led to affordable hand-held plenoptic cameras.
• Plenoptic cameras capture multiple views in a single shot using a micro-lens array.
• Joint processing of these views enables accurate depth map generation.

Purpose of the Study:

• To present a multi-resolution method for depth estimation from dense image arrays.
• To reduce computational complexity compared to global optimization methods.
• To achieve high accuracy in depth estimation while handling ambiguities and preserving edge details.

Main Methods:

• A local estimation approach based on maximizing total log-likelihood spatial density along epipolar lines.
• Utilizing epipolar plane images for local maximum likelihood estimation of the depth field.
• Implementing a multi-resolution scheme to reduce depth map resolution in ambiguous flat regions and preserve detail at edges.

Main Results:

• The proposed method significantly reduces computational complexity.
• High accuracy in estimated depth maps is achieved.
• The multi-resolution scheme effectively addresses accuracy losses in flat regions and preserves bandwidth at edges.

Conclusions:

• The developed system offers a favorable trade-off between accuracy, robustness, and handling of discontinuities.
• This method provides an efficient and accurate solution for depth estimation using plenoptic cameras.
• The multi-resolution strategy enhances the practical applicability of depth estimation techniques.