Efficient patch-based approach for compressive depth imaging.

Area of Science:

• Computer Vision
• Optical Engineering
• Image Processing

Background:

• Achieving extended depth of field in images is challenging with conventional cameras.
• Traditional methods often require multiple exposures or complex mechanical setups.

Purpose of the Study:

• To develop efficient camera hardware and algorithms for capturing extended depth of field (EDoF) images.
• To reconstruct all-in-focus images and depth maps from a single coded exposure.

Main Methods:

• Utilized a liquid lens to dynamically shift the focal plane.
• Employed a fixed binary mask modulated at different focal planes, synchronized with focal plane shifting.
• Developed reconstruction algorithms incorporating sparsity priors like group sparsity, tree structure, and dictionary learning.
• Leveraged patch-level operations for parallel computational structure.

Main Results:

• Demonstrated efficient reconstruction of all-in-focus images and depth maps from single coded exposures.
• Showcased the effectiveness of sparsity priors in enhancing reconstruction quality.
• Validated the system's performance through simulations and real-world experimental datasets.

Conclusions:

• The presented camera hardware and inversion algorithms effectively achieve extended depth of field imaging.
• The developed computational framework allows for efficient and parallelized image and depth reconstruction.