Extended depth of field (XDOF) used to mean stopping down a lens and accepting diffraction blur. Today, research prototypes replace the lens entirely with a coded mask or a diffractive element, then recover sharp images computationally across a far larger depth range. The shift matters because it decouples physical aperture size from depth performance — enabling thinner, lighter, and cheaper imaging systems for microscopy, robotics, and embedded vision.
How lensless cameras turn a mask into extended focus
A lensless camera places a coded amplitude or phase mask directly on the sensor. Each scene point spreads into a known pattern — the point spread function (PSF) — that the sensor records as a multiplexed measurement. Reconstruction solves an inverse problem to recover the scene. In conventional masks, the PSF changes with object distance, so a single calibration only works for a narrow depth slice. Radial masks avoid this: their PSF scales radially but keeps the same shape on the sensor, making it depth-invariant. Osaka University researchers showed that an optimized radial mask preserves this scaling property while improving frequency response, and they validated extended DOF on a prototype camera across continuous and dual-depth scenes [3]. The result is a single capture that reconstructs objects at widely different distances without mechanical focus.
Why the PSF shape matters more than its size
Depth-invariant PSFs let a system deblur with one spatially-invariant kernel instead of a depth-dependent set. Early diffusion-coded photography cameras demonstrated this principle: they extend DOF by designing a mask whose PSF stays similar across depths, then deconvolve with a single kernel [1]. Recovery quality hinges on two factors — the magnitude of the modulation transfer function (MTF) and how closely the actual PSFs match the design. If the PSF spectrum has deep nulls, noise amplification during deconvolution destroys detail. Optimizing the mask for a flat, broad MTF while keeping radial symmetry is therefore the central engineering tradeoff.
Snapshots add diffractive optics and deep learning
Microscopy demands speed. A 2025 ScienceDirect paper integrates a diffractive optical element (DOE) to modulate the optical field in a single exposure, then uses a deep network to enhance the reconstruction [2]. The DOE encodes depth information into the captured intensity pattern; the network learns to invert the combined optical-digital pipeline. This snapshot approach avoids multi-frame focus stacking, reducing motion artifacts and photon budget — critical for live-cell imaging. The same principle appears in a 2026 polarization-coded binary phase mask that exploits spherical aberration and Jones-matrix formalism to extend depth of focus [6]. Polarization adds a second encoding dimension, potentially improving conditioning of the inverse problem.
Computational pipelines that go beyond deconvolution
Pure deconvolution assumes a perfect shift-invariant model. Real systems need more. A 2023 Springer study frames EDOF as joint optimization of optics and processing: CodeV for lens design, global self-similarity priors for haze reduction, dark-channel-guided gradients for edge sharpening, and weighted grey-edge methods for color fidelity [7]. On benchmark metrics (PSNR, SSIM, RMSE) this combined pipeline outperforms conventional deconvolution alone. The lesson: mask design, calibration accuracy, and task-specific priors all contribute — none dominates in isolation.
What this means for your next imaging project
If you build embedded vision modules, lensless XDOF removes the focus actuator and reduces z-height — but you pay in compute (iterative reconstruction or neural inference) and calibration effort (measuring PSFs across depth and wavelength). For microscopy, snapshot DOE or polarization coding buys speed at the cost of added optical fabrication complexity. The Osaka radial-mask prototype demonstrates that a well-optimized amplitude mask can be fabricated with standard lithography and still deliver continuous depth coverage [3]. Choose the encoding strategy — radial amplitude, diffractive phase, polarization — based on your fabrication access, compute budget, and whether you need video-rate or still capture. The field is converging on co-design: mask pattern, calibration protocol, and reconstruction algorithm must be developed together.
Sources
- Extended Depth-of-Field Lensless Imaging - arXiv.org
- Diffusion coded photography for extended depth of field
- Snapshot extended depth-of-field microscopy via diffractive optical …
- Extended depth of focus using polarization-coded binary phase mask
- Extended depth of focus imaging using optics and image processing
- Extended Depth-of-Field Lensless Imaging Using an … - ResearchGate
- Super-resolution image projection over an extended depth of field using …
