SASSI - Super-Pixelated Adaptive Spatio-Spectral Imaging.

Area of Science:

• Optics and Photonics
• Computer Vision
• Image Processing

Background:

• Hyperspectral imaging traditionally faces trade-offs between spatial, temporal, and spectral resolutions.
• Achieving high-resolution hyperspectral video is computationally demanding and often requires specialized hardware.

Purpose of the Study:

• To introduce a novel video-rate hyperspectral imager with high spatial, temporal, and spectral resolutions.
• To demonstrate a scene-adaptive spatial sampling strategy for efficient hyperspectral data acquisition.

Main Methods:

• Acquisition of an RGB image to compute super-pixels, guiding spatial sampling for spectral measurements.
• Estimation of the hyperspectral image by fusing RGB data and spectral measurements using learnable guided filtering.
• Development of a low-complexity superpixel estimation for reduced computational overhead.

Main Results:

• The proposed technique enables high-quality hyperspectral reconstructions from scene-adaptive sampling.
• A lab prototype achieved hyperspectral video at 600x900 pixels, 10 nm spectral resolution, and 18 fps.
• The method offers significantly higher spatial and spectral resolutions compared to traditional snapshot hyperspectral cameras with minimal overhead.

Conclusions:

• The developed hyperspectral imager successfully captures high-resolution spectral video at high frame rates.
• Scene-adaptive spatial sampling guided by super-pixels is an effective strategy for efficient hyperspectral imaging.
• This approach overcomes previous limitations, paving the way for advanced applications in various scientific fields.