We give the gift of sight to machines. We do this by building computational imaging systems that tightly couple computer algorithms (CS) and imaging hardware (ECE). If successful, computational imaging has the potential to unleash an era of superhuman robotics.
Imagine if search-and-rescue robots could one day sense survivors through dense smoke.
Imagine if surgical robots could perform impossible surgeries by seeing details invisible to a human doctor.
To work towards these dreams, our lab focuses on three technical pillars that we push, explore and test.
Three Pillars of Our Lab
Pillar I: Computational Imaging
A computational imaging system tightly integrates software processing and optical capture to form novel imagery. Our group builds computational imaging systems that can see through scattering media, around corners, and more. We study visual constructs like light fields, time of flight imaging, polarization, and coherent optics. These constructs are parsed using signal processing and numerical methods. ECE 239 is a new UCLA course that covers the nascent field of computational imaging
Pillar II: Computational Robotic Imaging
In an approach we call “computational robotic imaging”, we generalize pillar I to a co-design of algorithms, optics and robotics. Inspired by the diversity of animal eyes, we believe that vision systems will evolve with the needs of robots. Imagine if a self-driving car had a vision system that could see obstacles before it turned a corner.
Pillar III: Computational Health Imaging
With an expertise in algorithms and optics, we build imaging systems for healthcare applications. Imagine the possibility of imaging through scattering media, like biological tissue. Ordinarily one has had to use X-rays or MRI. However, it might be possible to use novel optical and computational methods, like speckle or transient inversion to see deeper through the skin. The method at right is able to see deep subsurface veins (to identify blood clots), using spatially coded visible light. Imagine if similar techniques could be used as a low-cost, high throughput screening method for cancers.