Access Official Technical PDF

Share Summary:LinkedInEmail

NIR multispectral imaging can reveal information that a conventional visible-light camera does not capture. In a developed fundus camera, three near-infrared (NIR) bands work with a high-frame-rate CMOS imaging system to produce color-reconstructed images of the eye’s fundus. The system brings together optical filtering, compact sensor integration and image processing to address a practical challenge: capturing a clear image while the eye is moving.

The work also illustrates a broader principle in high-speed vision. Image quality depends not only on the sensor, but on coordinating illumination, spectral selection, frame rate and processing so that useful data can be collected and combined in real time.

NIR multispectral imaging for fundus observation

The fundus is the only location in the human body where arteries and capillaries can be observed directly and noninvasively. Imaging it can support examination of the retina, optic disc and blood vessels. The system described here was designed to make fundus capture possible without requiring an operator to position and operate a conventional camera for the user.

Its image sensor uses six spectral channels: red, green and blue, alongside NIR1, NIR2 and NIR3. The NIR bands are approximately 780–800 nm, 870 nm and 940 nm. Silicon sensors can detect light into the near-infrared range, but conventional digital cameras typically use an NIR-cut filter to produce visible-light color images. This design instead separates selected NIR signals and uses them to reconstruct a color image.

Interpolation and color-correction processing combine the three NIR channels. The paper reports that the system can reproduce color images using NIR signals in an environment with no visible light (0 lux). This is a processed representation of multispectral data, rather than a conventional visible-light photograph.

How the Fabry–Perot filter selects wavelengths

To distinguish the NIR bands at the sensor, the camera uses multilayer interference band-pass filters based on the Fabry–Perot principle. Alternating films with different refractive indexes form reflective structures. A spacer layer between them creates a resonant cavity that transmits light near a selected wavelength while reflecting much of the light outside that band.

Changing the spacer’s thickness shifts the filter’s transmission peak: reducing the thickness moves the peak toward shorter wavelengths, while increasing it moves the peak toward longer wavelengths. This makes it possible to create different NIR passbands by varying the central high-refractive-index layer while keeping the surrounding layer structure consistent.

The filters are fabricated as a patterned array on a CMOS image sensor. The reported multilayer stack is approximately 1.1 µm thick, with a measured example around 1.18 µm. Its thin profile supports integration with fine-pitch pixels; the paper describes applicability to pixels with an area below 2.25 µm², equivalent to a 1.5 µm pixel pitch. Fabrication includes forming and patterning the central layer at different thicknesses for the three NIR bands, then completing the shared upper layer structure.

High-speed tracking addresses eye movement

Even when a person tries to hold their gaze steady, the eye makes small, involuntary movements. The paper notes movement at approximately 100 Hz, which can blur images or make continuous capture difficult. A high-speed camera operating at approximately 1,000 frames per second can sample this motion at a much higher rate than a conventional still-image capture approach.

High frame rate alone does not create a stable image. The system applies image tracking to follow the fundus across frames. The described method uses template matching with a planar projective transformation and nonlinear minimization. Tracking allows the processing system to compensate for movement and select or align image information from successive frames.

Multiple frames can then be integrated to reduce image noise. In practical terms, the approach combines two functions: tracking helps maintain a consistent view of the fundus, while frame accumulation improves the signal-to-noise ratio of the resulting image. The reported outcome is a stable, lower-noise fundus image despite the eye’s inherent motion.

Illumination and user operation

The optical arrangement directs NIR illumination through a fundus lens toward the pupil using a Maxwellian illumination configuration. Reflected light from the fundus passes through the imaging optics and a beam splitter to the camera. The design is intended to let individuals capture and check their own fundus images without the dazzling associated with the strong flash used in some high-speed imaging setups.

That capability is significant as a system design objective, but the camera should not be interpreted as a standalone diagnostic tool. The paper presents potential use in health monitoring and the examination of retinal conditions; clinical diagnosis and interpretation remain matters for qualified healthcare professionals.

What the design demonstrates

This camera integrates several engineering elements that are often treated separately:

  • Spectral filtering: Fabry–Perot multilayers isolate three NIR wavelength bands on a CMOS sensor.
  • Compact integration: A thin patterned filter structure supports use with fine-pitch image pixels.
  • Motion handling: High-frame-rate acquisition and image tracking address involuntary eye movement.
  • Image combination: NIR interpolation, color correction and multi-frame integration produce color-reconstructed, lower-noise images.

For engineers working on intelligent imaging systems, the practical lesson is that sensor performance is only one part of the solution. The filter design determines which spectral information reaches each pixel; illumination shapes the optical signal; and processing must account for both motion and noise. Coordinating these choices can expand what a compact camera is able to observe.

Conclusion

The NIR multispectral fundus camera pairs three near-infrared bands with a thin Fabry–Perot filter array and high-speed image tracking. Its approach shows how wavelength selection, sensor fabrication and frame-based processing can work together to capture stable images under motion. Although the system’s proposed health applications require appropriate clinical interpretation, its core engineering principles are relevant to the development of compact, real-time imaging systems that need to extract useful information from fast-changing scenes.

Access Official Technical PDF

About The Author