Small Image Sensors and Big Visual Data
90 views · Published 27 July 2016 · 1:05:06 · Indexed 6 November 2026
Channel: Microsoft Research · 2016 · Science & Technology
This talk will discuss small image sensor designs and big visual data parsing and their applications. We live in a technology world in which cameras are shrinking in size and photos and videos are exploding in quantity. My research has been dealing with technical challenges in both areas. I will show a single-chip design for compressive sensing of high-speed high-dynamic-range videos. The design has the benefit of achieving nearly 100% light throughput and uses reduced sampling rate to reduce power consumption. I will show how to enhance low light performance and remove hand shaking in video capture using a portable array of small cameras. I will show a nonparametric approach to the image parsing problem applied to street view images and face images; that is, classify each pixel into sky, road, building, eye brows, lips, etc, given an input image. I will show the potential applications of image parsing in touchpad-based face image retrieval and privacy protection in video surveillance.