Computer Vision News - April 2021

16 Congrats Doctor! Historically, cameras have been optimized to capture high quality photographs on film with no post-processing. However, today’s images are almost entirely digital, and post-processing is ubiquitous. This raises the question: Is the current camera architecture best for all applications? For instance, lenses are an integral part of most imaging systems, but they are typically heavy, expensive, and do a poor job of recording depth information. With my collaborators*, we replaced the lens of a traditional camera with a diffuser -- a bumpy piece of thin plastic that refracts light -- to create a compact lensless camera which we call DiffuserCam . The diffuser scrambles the incoming light making the sensor measurement unintelligible to a human, but we developed an algorithm to recover the image from the raw data using a physically-based model of the system. Grace Kuo recently completed her PhD at the University of California, Berkeley , where she worked on designing computational cameras, microscopes, and displays . By jointly designing the optics and processing algorithms, she demonstrated imaging systems with simplified hardware (for example, no lenses) and expanded capabilities (for example, 3D). After graduating, Grace began working at Facebook Reality Labs where she develops new display technology for virtual and augmented reality applications. Congrats, Doctor Grace! In addition, our lensless architecture naturally records 3D information . By applying techniques from compressed sensing, we demonstrated recovery of an entire 3D volume from a single 2D measurement taken with DiffuserCam .