Jacobs School of Engineering, UC San Diego

Photonic Systems Integration




Platform Motion Blur Image Restoration System

  help help help help  

Compressive imagers acquire images, or other optical scene information, by a series of spatially ltered intensity measurements, where the total number of measurements required depends on the desired image quality. Compressive imaging (CI) o ers a versatile approach to optical sensing which can improve SWaP for hyperspectral imaging or feature-based optical sensing. Here we report the rst (to our knowledge) systematic performance comparison of a CI system to a conventional focal plane imager for binary, grayscale, and natural light (visible color and infrared) scenes. We generate 1024  1024 images from a range of measurements (0.1% to 100%) made using digital (Hadamard), grayscale (Discrete Cosine Transform) and random (noiselet) CI basis sets, and for varying numbers of measurements. Comparing the outcome of the compressive images to conventionally acquired images, each made using 1% of full sampling, we conclude that the Hadamard Transform o ered the best performance and yielding images with comparable aesthetic quality and slightly higher spatial resolution than conventionally acquired images.

Schematic of system

Prototype system

Transforms used in characterization of Single Pixel Camera

Error characteristics

Comparison Results
Canon SLR image 1% data (top),
Canon SLR image 1% data upsampled (center),
Compressive Image made using Hadamard transform (bottom)

System Performance Parameters


S. Olivas, Y. Rachlin, L. Gu, B. Gardiner, R. Dawson, J. Laine, and J. Ford, "Characterization of a compressive imaging system using laboratory and natural light scenes," Appl. Opt. 52, 4515-4526 (2013).

S. Olivas, Y. Rachlin, L. Gu, B. Gardiner, R. Dawson, J. Laine, and J. Ford, "Single Pixel Compressive Imaging of Laboratory and Natural Light Scenes," (COSI), OSA, CTu1C.2 (2013). For presentation, click here.