Computational Imaging

Computational imaging unites bespoke optical hardware with advanced reconstruction algorithms to transcend the limits of conventional cameras. Rather than forming images directly on a pixel array, these systems encode scene information—spatial, spectral, temporal or polarimetric—into indirect measurements via structured illumination, coded apertures or programmable metasurfaces. Single-pixel and ghost-imaging techniques exploit correlations between known patterns and bucket-detector responses, while compressive-sensing frameworks recover high-resolution or hyperspectral data from far fewer measurements by leveraging signal sparsity. The integration of machine-learning refines these reconstructions, suppressing noise and artefacts under extreme conditions such as low light, sub-Nyquist sampling or photon-starved regimes. From terahertz security screening to mid-infrared microscopy and four-dimensional light-field capture, computational imaging is enabling new applications that lie beyond the capabilities of traditional optics.

Research from Nature Portfolio

Recent advances have realised mid-infrared, single-photon computational imaging by imprinting time-varying patterns onto the infrared field and converting the light to visible wavelengths via nonlinear sum-frequency generation. Coupled with compressed-sensing and deep-learning reconstructions, this scheme achieves room-temperature, single-photon sensitivity under sub-Nyquist sampling. Another seminal contribution employs programmable metasurfaces as large-aperture panels for millimetre-wave imaging; by sweeping frequency and generating spatially diverse field patterns, inverse-scattering algorithms reconstruct diffraction-limited images of human-scale scenes without mechanical scanning. Complementing these hardware innovations, the concept of semantic regularisation has been introduced for electromagnetic inverse problems: by embedding language-derived priors into microwave reconstructions, it becomes possible to conceal selected subjects or transform their postures, advancing privacy-aware touchless interaction.

Research from all publishers

A cascaded compressed-sensing single-pixel camera has been developed to address high-dimensional optical imaging. This multistage architecture sequentially exploits sparsity across spatial, spectral and temporal domains, enabling tunable hyperspectral LIDAR and efficient four-dimensional scene capture. In the temporal domain, a mid-infrared ghost-imaging protocol transfers near-infrared modulation patterns to an idler beam via difference-frequency generation, facilitating ultrafast pump–probe measurements in spectral regions lacking fast modulators. A comprehensive review of single-pixel imaging highlights the evolution of mask design, adaptive sampling strategies and reconstruction algorithms—spanning compressive sensing, iterative optimisation and data-driven methods—that have extended applications from terahertz microscopy to real-time three-dimensional video.

Computational Imaging publication trend

The graph below shows the total number of articles in computational imaging across all publications each year (not limited to Nature Index journals).

Technical terms

Single-pixel camera: A system that projects sequential spatial or spectral patterns onto a scene and records the total transmitted or reflected light with a single detector, reconstructing images by correlating measurements with known patterns.

Ghost imaging: An indirect imaging approach that reconstructs a scene by correlating bucket-detector signals with structured illumination patterns, without requiring a pixelated sensor on the object-interacting field.

Compressed sensing: A signal-processing paradigm that recovers high-fidelity images from undersampled data by exploiting sparsity in a transform domain, thereby reducing measurement requirements.

Programmable metasurface: A planar array of electronically or optically tunable subwavelength elements that shapes electromagnetic wavefronts in amplitude, phase and polarisation for coded sampling.

Semantic regularisation: The incorporation of high-level, language-derived prior knowledge into inverse-problem solvers to guide reconstructions and enforce context-aware image semantics.

References

1. Mid-infrared single-pixel imaging at the single-photon level. Nature Communications (2023).
2. Large Metasurface Aperture for Millimeter Wave Computational Imaging at the Human-Scale. Scientific Reports (2017).
3. Semantic regularization of electromagnetic inverse problems. Nature Communications (2024).
4. Cascaded compressed-sensing single-pixel camera for high-dimensional optical imaging. PhotoniX (2024).
5. Mid-infrared computational temporal ghost imaging. Light: Science & Applications (2024).
6. Single-pixel imaging 12 years on: a review.. Optics Express (2020).