Extending Depth of Field via Depth Invariant Imaging

36 concepts26 papers1 proposal

In traditional optics, photographers and optical engineers are bound by a frustratingly rigid trade-off: the conflict between depth of field (DOF) and light-gathering power. If you open a lens's aperture wide to let in more light or achieve high lateral resolution, the depth of focus shrinks to a razor-thin plane. Anything slightly in front of or behind this plane becomes a blurry slurry of light.

Computationally reversing this blur is notoriously difficult because standard defocus is depth-dependent. An algorithm cannot easily sharpen an image without first knowing exactly how far away each object is, creating a challenging chicken-and-egg problem.

To break this bottleneck, a powerful paradigm has emerged: depth-invariant imaging. Instead of attempting to capture a perfectly sharp physical image, these systems intentionally engineer the optics or the capture process so that the blur characteristic—the Point Spread Function (PSF)—remains identical across a massive range of depths. Because the blur is uniform and preserves high-frequency details, a single, depth-independent computational step can restore perfect sharpness across the entire volume.


Temporal Coding: The Magic of Focal Sweep

The most intuitive way to understand depth invariance is to look at how we can code an image over time during a single exposure.

Imagine a standard camera focusing on a scene. If an object sits exactly at the focal plane, its light converges to a sharp point on the sensor. If it moves away, its light spreads into a wide blur circle.

Now, consider what happens if we physically translate the image sensor along the optical axis at a constant velocity during a single exposure. This technique, known as focal sweep, completely changes the math. As demonstrated by Nagahara et al. 2008, every object point in the scene—regardless of its distance from the camera—experiences the exact same sequence of in-focus and out-of-focus states during the sweep.

Static Sensor:  [Object Depth A] ---> Sharp Point
                [Object Depth B] ---> Wide Blur Circle (Depth-Dependent)

Focal Sweep:    [Object Depth A] ---> Sweeps through focus ---> Integrated PSF
                [Object Depth B] ---> Sweeps through focus ---> Identical Integrated PSF

Because the sensor integrates the light over the entire exposure, the resulting composite blur kernel becomes completely invariant to depth. As Hajime Nagahara notes in a comprehensive review of computational photography (Nagahara 2019), this motion-based coding preserves the signal-to-noise ratio (SNR) far better than physical masks because it never obstructs or discards incoming light.

Once the exposure is complete, a single, uniform deconvolution operation can be applied to the entire image to peel back the blur, revealing an all-focus image (Nagahara 2013). This concept has been successfully adapted to complex optical setups; for instance, Yokoya et al. 2015 integrated a sensor sweep into catadioptric (curved mirror) panoramic systems, achieving a 30-fold extension of focus while maintaining a wide aperture in low-light environments.


Wavefront Engineering: Non-Diffracting Needle Beams

While moving a sensor works beautifully for macro photography and computer vision, mechanical translation is often impractical for high-speed microscopy or laser surgery. In these domains, researchers achieve depth invariance by engineering the wavefront of the light itself, transforming standard light cones into non-diffracting beams.

A typical laser beam follows a Gaussian profile: it narrows to a tight waist at its focus and immediately diverges. If you use a Gaussian beam for imaging, your high-resolution zone is incredibly shallow.

By inserting conical optical elements called axicons or specialized metasurfaces into the optical path, a Gaussian beam can be transformed into a quasi-Bessel beam. As explained by Wang et al. 2020, a Bessel beam does not obey standard diffractive spreading; instead, its phase profile causes the light to continuously interfere with itself along the axis of propagation. This creates a narrow, highly concentrated "needle" of light that maintains its intensity and tight lateral focus over a long distance.

Gaussian Beam:       \       /
                      \_____/   <-- Shallow focal waist
                      /     \
                     /       \

Bessel/Needle Beam:  ============================= <-- Elongated, depth-invariant focus

This elongated focal zone provides an elegant solution for imaging thick, irregular, or dynamic volumes without mechanical z-scanning:

  • Volumetric Biological Imaging: Yang et al. 2016 utilized Bessel beams in two-photon stereomicroscopy to capture substantial volumes in a single pass, drastically accelerating the visualization of fast biological events. Similarly, Glandorf et al. 2024 leveraged Bessel-beam optical coherence microscopy to map the intricate, multi-scale 3D morphology of cerebrovascular networks without vertical scanning.
  • Slide-Free Histology: To bypass the tedious process of slicing tissue samples, Cao et al. 2024 transformed ultraviolet laser foci into needle-shaped profiles, allowing parallel photoacoustic microscopes to clearly image highly uneven, fresh tissue surfaces.
  • Industrial and Medical Sensing: In terahertz sensing, Nowack et al. 2023 demonstrated that Bessel beams preserve measurement precision even when a sample's axial position fluctuates. In automated laser bone surgery, Hamidi et al. 2024 showed that Bessel-like profiles maintain image clarity and ablation efficiency across varying target distances, eliminating the need for real-time mechanical refocusing.

Spatial Multiplexing: Heterogeneous Focal Assignment

A third facet of this research direction relies on spatial multiplexing—distributing different focal properties across an aperture, a lens array, or a mask so that all depths are sampled simultaneously.

Instead of forcing a single lens to be depth-invariant, systems can utilize a heterogeneous distribution of focal lengths. For example, Koike et al. 2010 developed a light-field capture array using liquid lenses arranged in a structured checkerboard pattern of alternating focal lengths. Because different lenses are tuned to different depths, the array collectively captures sharp high-frequency data across an extended synthetic depth of field.

Taking this further, the "Fourier DiffuserScope" developed by Liu et al. 2020 places a diffuser containing randomly distributed microlenses of varying focal lengths directly into a microscope's pupil plane. This ensures that every depth plane in the specimen volume is brought into focus by a distinct subset of microlenses, mitigating the severe resolution degradation typically seen away from the native focal plane in standard light-field microscopy.

Spatial multiplexing can also be achieved through flat optics and computational masks:

  • Quadratic Metalenses: Lassalle et al. 2021 engineered an ultra-compact metalens with a unique quadratic phase profile. This profile deliberately elongates the focal spot along the optical axis, expanding the depth of focus for compact fingerprint sensors.
  • Optimized Radial Masks: In lensless imaging, where there is no physical lens to focus light, Silva Neto et al. 2023 demonstrated that optimizing the frequency response of a radially symmetric coded mask produces a highly stable, depth-invariant response, allowing sharp image reconstruction across a wide depth range.
  • Active Diaphragms: Rather than moving the sensor or using static masks, Xiao et al. 2022 positioned an adjustable liquid crystal lens directly at the system diaphragm. By capturing images during the lens's rapid transient state, they acquired multi-focus stacks without changing the system's magnification or field of view, completely eliminating the "lens breathing" that plagues traditional depth-from-focus algorithms.

Finally, this expanded depth tolerance has proven vital for industrial metrology. In phase-measuring deflectometry—a technique used to map micro-scale surface defects—traditional microscopes struggle to keep both the projected reference patterns and the uneven test surface in focus simultaneously. By combining pinhole camera models with structured sinusoidal illumination (Hu et al. 2023) or introducing concave focusing mirrors (Gao et al. 2023), researchers have expanded the depth of field by up to 20 times, allowing for rapid, non-planar quality control without constant mechanical adjustments.

Summary

Whether by sweeping a sensor through time, shaping light into non-diffracting needles, or scattering focal lengths across space, depth-invariant imaging works because it rethinks the role of optics. Instead of demanding that a lens deliver a perfect image directly to our eyes, these techniques ensure that high-frequency information is safely encoded in a uniform, predictable manner, leaving the final act of focusing to computation.

Key Foundational Work

Flexible Depth of Field Photography (2008) — contributed 2 concept elements to this direction.
Three-Dimensional Image Information Media: Optical and Synthetic Focusing Using Liquid Lens Array (2010) — contributed 2 concept elements to this direction.
History & citationVersion 1 · 2026-10-04

Current · Permanent link

View frozen version 1

Cite this direction

Latest version (a living page):

@misc{lacuna_axial_sensor_translation_682ebad8,
  title = {axial sensor translation},
  author = {{Lacuna}},
  howpublished = {Lacuna research direction},
  year = {2026},
  url = {https://lacuna.tiptreesystems.com/direction/axial-sensor-translation/rdr_682ebad8cdbb48f4af6ce4b4e755380b},
  note = {Latest version of a living research direction; cite a version for a fixed text. Accessed 2026-11-19},
}

This exact version (version 1, frozen text):

@misc{lacuna_axial_sensor_translation_682ebad8_v1,
  title = {axial sensor translation},
  author = {{Lacuna}},
  howpublished = {Lacuna research direction},
  year = {2026},
  url = {https://lacuna.tiptreesystems.com/direction/axial-sensor-translation/rdr_682ebad8cdbb48f4af6ce4b4e755380b/v/1},
  note = {Version 1, frozen 2026-10-04. Accessed 2026-11-19},
}
About

These directions overlap with this one through shared papers.

See something wrong? Report a mistake