Snapshot Hyperspectral Imaging with a Computational Lensless Camera

Taeyoung Kim, Seung Ah Lee* 

Hyperspectral imaging (HSI) collects spatio-spectral information of objects, useful in a wide range of applications including biomedical imaging. However, most HSI systems require multiple measurements or complex hardware arrangements.

We propose a compact snapshot HSI system, composed only of a monochromatic CMOS image sensor, a transparent phase-mask, and a linear variable filter (LVF). By leveraging the multiplexing capabilities of the lensless imaging, our system effectively achieves snapshot HSI with simple hardware.

(a) The LVF filters specific wavelengths of light depending on position along the filter. By adding an LVF in our lensless camera, different wavelength components of the light from a white-light point source pass through the different parts of the phase mask and form a wavelength-dependent PSF (spectral PSF). 

(b) We can formulate the sensor’s measurement as a summation of convolutions between the spectral PSF and the object’s image at the specific wavelengths. Minimizing the difference between the sensor measurement and the convolution model with an optimization iteratively, we can recover the hyperspectral data stack.

For calibration, we measured a standard ColorChecker target with a spectrometer and computed for calibration functions to transform the reconstructed image to the reflection spectrum. As a result, we can get multiple spectral images along 416~933 nm wavelengths from a  single measurement with a monochromatic image sensor.

Our system is capable of capturing snapshot HSI with lensless image reconstruction, with compact and cost-effective hardware compared to conventional HS cameras. While most HS image sensors with a finite number of spectral filters patterned on the pixels have a direct trade-off between the spectral and the spatial resolution, computational imaging with lensless cameras can potentially decouple this trade-off, using advanced reconstruction algorithms with proper regularization.

References :