Broadband Obstruction-Free Imaging

• Broadband obstruction-free imaging is a multidisciplinary field combining diffractive and meta-optics with computational reconstruction to achieve clear images despite occlusions.
• Recent advancements leverage achromatic diffractive optics and split-spectrum filtering to correct chromatic aberrations and suppress near-field obstructions.
• Emerging computational strategies such as end-to-end learning and iterative phase retrieval enable high-fidelity image recovery from visible to extreme-ultraviolet wavelengths.

Broadband obstruction-free imaging refers to the optical and computational strategies enabling high-resolution, full-field imaging over large spectral bandwidths even in the presence of occlusions, scattering, or conventional lens limitations, without physical obstructions or central obscurations in the optical path. This domain synthesizes advances in diffractive and meta-optics, lensless imaging, low-coherence illumination, compressive measurements, and computational reconstruction to provide clear images across strongly chromatic, obstructed, or scattering environments. The field addresses longstanding challenges in achieving simultaneously compact, planar, and high-fidelity imaging across broad wavelength ranges, with utility in visible, near-infrared, and even extreme-ultraviolet regimes.

1. Physical Principles and Device Architectures

Broadband obstruction-free imaging leverages both engineered optics and the underlying wave physics to enable imaging through scattering layers, occlusions, or over extended bandwidths.

• Achromatic Diffractive Optics: Planar diffractive lenses (ADLs) implement phase patterns $φ(r,λ) = \frac{2π}{λ}[n(λ)-1] h(r)$, using multi-level relief to achieve phase delays invariant across wide spectral bands. Concentric zones with discrete heights (256 gray-levels, as small as 0.3–3 μm features) allow for efficient compensation of chromatic focal shifts (Mohammad et al., 2017).
• Meta-optics and Metalenses: Engineered subwavelength structures, such as geometric-phase metalenses fabricated in SiN, utilize meta-atom orientation to impart precise broadband phase profiles (e.g., $φ(x,y)=2θ(x,y)$). These can further exploit depth–wavelength symmetries to manipulate focus and defocus properties, central to recent split-spectrum, obstruction-removing designs (Yoon et al., 27 Jan 2026, Fröch et al., 2024).
• Lensless and Coded Approaches: Strategies dispensing with conventional imaging optics include:
  • Two-pulse, spectrally-encoded Fresnel/diffraction imaging, reconstructing images by exploiting time-delay and multi-wavelength amplitude constraints without spectral filtering or supports (Witte et al., 2013).
  • Single-broadband-antenna imaging with strongly disordered media, which encodes scene information via frequency-diverse, spatially randomized Green's functions, enabling far-field, subwavelength reconstructions by compressed sensing (Li et al., 2014).

Devices achieving true unobstructed imaging avoid central obscurations, spider vanes, or relay optics—offering planar or nearly flush apertures for direct sensor integration (Mohammad et al., 2017, Fröch et al., 2024).

2. Chromatic Correction and Bandwidth Optimization

Classic diffractive and meta-optical designs are limited by severe chromatic aberrations, as the focal length and point-spread function (PSF) scale with wavelength ($f ∝ 1/λ$). Recent approaches achieve broadband performance through:

• Multi-Spectral Optimization: Incorporating multiple design wavelengths (e.g., $\lambda_i = 450,500,...,750$ nm) in the phase profile optimization, maximizing average focusing efficiency over the target band (Mohammad et al., 2017).
• EDOF Phase Engineering: Adding polynomial or numerically optimized terms to the phase to equalize the PSF across wavelengths, resulting in extended depth-of-focus (EDOF) or spectrally invariant blur (Fröch et al., 2024).
• Split-Spectrum Filtering: Introducing wavelength-pass and stop-bands tailored to decouple the depth–wavelength coupling, enabling suppression of sharply focused obstructions near the lens while preserving broadband, in-focus imaging of distant scenes (Yoon et al., 27 Jan 2026).

Simulations and experimental designs confirm the possibility of achieving diffraction-limited or near-diffraction-limited resolution (e.g., $d = 0.61λ/NA$), with focusing efficiency $\eta$ exceeding 40% across 300 nm bandwidth for scalar diffractive designs, and $>34\%$ simulated over similar bands at NA = 0.9 for metasurfaces (Mohammad et al., 2017).

3. Obstruction and Scattering Mitigation

A critical feature is maintaining image quality in the presence of occlusions (e.g., dust, rain, fences) or through turbid/scattering media.

• Metalens-Based Obstruction Filtering: The depth–wavelength symmetry in diffractive optics results in both distant objects and near obstructions being in focus at different wavelengths. By applying a split-spectrum filter (only transmitting a tailored $\Lambda_{\text{pass}}$ for each color channel), near-depth obstructions are spectrally rejected while far-scene sharpness is preserved. Post-hoc neural refinement then restores residual detail without hallucination (Yoon et al., 27 Jan 2026).
• Memory-Effect Speckle and R-Autocorrelation: In thin-scattering scenarios, the angular optical memory effect results in speckle patterns with spatially shift-invariant PSFs, enabling single-shot, broadband autocorrelation-based reconstruction. The R-autocorrelation technique aligns and averages bright speckle sub-regions to suppress background, achieving contrast-to-noise ratios (CNR) up to $7.8$ in broadband white-light, far surpassing classical autocorrelation ($\text{CNR}<2$ under $\Deltaλ>100$ nm), with SSIM up to $0.85$ (Wu et al., 2018).
• Low-Coherence, Multimode Illumination: Broadband, low-spatial-coherence fiber ASE sources (e.g., $λ=1055$ nm, $\Deltaλ=74$ nm, $\langle M\rangle=50$ modes) generate multiple mutually incoherent speckle patterns, enabling spatial and spectral compounding to average speckle contrast down to $C<0.02$, well below human detectability (Redding et al., 2015). This yields speckle-free, full-field imaging behind static diffusers.

4. Computational and Algorithmic Strategies

Powerful inversion, deconvolution, and learning-based approaches are integrated with novel optics to realize broadband obstruction-free imaging:

• Direct-Binary-Search and Gradient-Based Inverse Design: Optimization of multi-level phase profiles or metasurface orientation maps is performed by maximizing empirical figures of merit (focusing efficiency, average MTF), using direct-binary-search for scalar diffractive optics (Mohammad et al., 2017) and automatic differentiation for meta-optics (Fröch et al., 2024).
• End-to-End Learning with Dataset Pairing: Joint optimization of the meta-optic PSF and a computational reconstruction network is performed using paired captures—meta-optic and compound-lens images—enabling structural similarity SSIM up to 0.85 even with strong chromatic blur (Fröch et al., 2024).
• Lensless Iterative Phase Retrieval: Algorithms enforcing multi-wavelength amplitude constraints at the detector plane are used to reconstruct both amplitude and phase of the object in lensless, broadband Fresnel diffraction geometries, converging robustly even under high noise and large spectral diversity (Witte et al., 2013).
• Sparse Recovery from Broadband Compressive Measurements: In strongly disordered media, the measurement matrix assembled from frequency-diverse single-antenna responses mimics incoherent random projections, permitting $\ell_1$-minimization (basis pursuit denoise) to achieve subwavelength recovery ($\sim λ/10$) in the far field (Li et al., 2014).

Neural networks with U-Net or encoder–decoder architectures (including diffusion models) are pivotal in inverting extended PSF, correcting complex aberrations, and enhancing raw physical outputs for both obstruction-free and chromatically uniform image recovery (Fröch et al., 2024, Yoon et al., 27 Jan 2026).

5. Metrics, Quantitative Performance, and Benchmarks

Quantitative reporting in state-of-the-art systems includes:

System/Paper	Key Metric	Reported Value
Planar ADL (Mohammad et al., 2017)	Focusing efficiency (NA=0.05)	40–45% (450–750 nm)
Planar ADL (Mohammad et al., 2017)	Diffraction-limited FWHM	Within 5% of $0.61λ/NA$
Broadband metalens (Fröch et al., 2024)	MTF @ 70/100 lp/mm (post-recon)	≥0.5/0.3 over 480–680 nm
R-autocorrelation (Wu et al., 2018)	Peak-to-background (broadband)	6.7 ($\times$3.3 over classic)
Fiber ASE imaging (Redding et al., 2015)	Speckle contrast, spatial+spectral	$C<0.02$ (below visibility)
Split-spectrum metalens (Yoon et al., 27 Jan 2026)	PSNR improvement (obstructed)	+32.29% over hyperbolic
Split-spectrum metalens (Yoon et al., 27 Jan 2026)	mAP/IoU/mIoU (object/segmentation)	+13.54/48.45/20.35%

Fidelity is further confirmed via USAF-1951 target readings, structural similarity indices (SSIM), and task-specific gains (object detection, semantic segmentation).

6. Implementation Guidelines and Practical Limitations

Translation to practical, scalable imaging systems involves:

• Fabrication: Planar ADLs and metasurfaces are fabricated via grayscale mask lithography, e-beam lithography, or nanoimprint, with feature tolerances established by simulation (e.g., zone-height error $σ_h ≈ 150$ nm reduces $\eta$ by 10–20%) (Mohammad et al., 2017).
• Integration: Thin-form-factor, obstruction-free lenses are paired directly with CMOS/CCD sensors. For robust mechanical integration, mountings accommodate focal lengths on the mm to few-μm scale (Mohammad et al., 2017, Fröch et al., 2024).
• Spectral Range and Source Engineering: Appropriate filters (IR/UV-block) or rare-earth dopant changes adapt operation to specific bands. ASE fibers allow scaling of power, bandwidth, and mode count for application-specific needs (Redding et al., 2015).
• Computation and SNR Trade-offs: For R-autocorrelation and lensless phase retrieval, computational burden grows with sub-region count and phase-retrieval iterations ($\sim$20 s per image), and many frames are needed for high-noise, ultra-broadband sources (Wu et al., 2018, Witte et al., 2013).
• Limitations: Objects must reside within the isoplanatic or memory-effect window in scattering systems. Born-approximation-based compressive methods require linear or weak-scatterer regimes (Li et al., 2014). High-NA meta-optics may experience angle-dependent color shifts and efficiency loss at extreme off-axis incidence (Fröch et al., 2024). Scaling to $\gtrsim$cm-aperture meta-optics intensifies design and compute demands.

7. Perspectives and Research Directions

Recent advances demonstrate that the longstanding trade-off between bandwidth, planar form factor, and image quality can be overcome by carefully co-designing optical phase, spectral filtering, illumination properties, and computational inversion frameworks (Fröch et al., 2024, Yoon et al., 27 Jan 2026). Compact, high-throughput, and obstruction-free modules are now deployable in space-constrained machine vision, drones, endoscopy, and mobile sensing. Further scaling—toward larger apertures and shorter wavelengths (X-ray or EUV)—will require improved fabrication, robust phase retrieval, and rapid computational inversion.

A plausible implication is that as learning-based and quantitative phase optimization techniques mature, the distinction between optical and computational correction will blur, with jointly optimized devices surpassing classical optics in both performance and compactness. Bridging physical-layer signal conditioning (e.g., depth-locked PSF engineering, speckle decorrelation) with robust, real-time machine learning will define the future of full-spectrum, obstruction-free imaging platforms.