Projects‎ > ‎

Haze Removal via Optimal Transmission Map


[T08] R. Tan, “Visibility in Bad Weather from a Single Image,” in Proc. IEEE Conf. Computer Vision and Pattern Recognition, pp.1–8, June 2008.

This work can derive a very accurate transmission map for single image dehazing. The superiority comes from:
  • We leverage the closed-form image matting model; and
  • We formulate it as a quadratic programming (QP) optimization prob. with global optimum.
To address the computational burden, recently we develop a multi-scale framework to accelerate cpu time by 10 times. The basic idea is 
  • To reduce # unknowns in QP from coarse to fine (illustrated in the right figure). 
  • Some locations, which show small gradient responses in the coarse-scale transmission, are simply estimated by interpolation.

The gradient responses of transmission map (x-axis) vs. probability (y-axis). We assume it follows Gaussian distribution and in different scales, we control the selected range for update by its mean and standard derivation .