Wednesday, April 2, 2008
Neyman Seminar, Evans 1011 | 1011 Evans Hall

Speaker: Raanan Fatal, Miller fellow, Mathematics, U.C. Berkeley

Title: Single Image Dehazing using de-correlation

In this talk I'll present a new method for estimating the optical
transmission in hazy scenes given a single input image. Based
on this estimation we eliminate the scattered light to increase the
visibility of the scene and recover haze-free contrasts. In this new
approach we formulate a refined image formation model that accounts
for surface shading in addition to the transmission function. This
addition allows us to resolve ambiguities in the data by searching
for a solution in which the resulting shading and transmission
components are statistically uncorrelated. We use this principle
to estimate automatically the color of the haze. We show that our
method can significantly remove the haze layer and produce reliable
transmission estimation from which, in certain cases, we extract
scene depths and use them for other applications such novel view
synthesis and image refocusing.