Computer Vision Talks at Columbia University

Rendering with approximate concentric mosaics, and Textureless calibration

Sing Bing Kang

Microsoft Research,

Seattle, WA.

Host: Shree K. Nayar

11:30 a.m. – July 17^th, 2000  

CAVE (Computer Vision Lab) 6^th floor CEPSR, Schapiro Building

Abstract

This is a two-part talk. The first part is based on work done recently at Microsoft Research, in collaboration with Jinxiang Chai (CMU) and Harry Shum (Microsoft Research China). It is an offshoot of Harry Shum and Li-Wei He's concentric mosaic work published in SIGGRAPH'99; it explores a more practical means for building and rendering concentric mosaics. Here, images are captured using only a hand-held camera. The key is the use of the signed Hough space to bin the captured rays. Various kinds of bilinear interpolation schemes can be used in this space for rendering, and I'll show some results involving real scenes.

The second part of the talk is based on work done while I was with Compaq's Cambridge Research Lab, in collaboration with Richard Weiss. I'll describe how to perform the seemingly impossible task of calibrating a camera using just a flat, textureless Lambertian surface under uniform illumination.