Embedded Files
under construction ...
Image De-Hazing for Radiance, Airlight and Transmission
}Dichromatic atmospheric scattering model: I(x) = t(x)J(x)+(1-t(x))A;
}J(x): surface radiance vector (haze free);
}A: constant airlight color vector (veiling light);
}t(x): medium transmission along the ray.
}t(x)=exp(-r*d(x)) with d(x) as depth and r as scattering coeff.
}The incoming light blended with the airlight (by atmospheric particles);
}Haze induces its visual effect is blurring of distant objects ( a clue for depth inference, i.e. aerial perspective);
}The loss of contrasts and colors due to haze or fog as image averaging with a constant color A.
}De-hazing or de-foggy is ill-posed too!
}multiple images (polarization) under different lighting [Joshi & Cohoen, 2010] or additional information (depth or geometry) [Kopf et al.’2008];
}Assumptions or strong prior, esp. for single image.
}De-noise and de-haze simultaneously or sequentially.
}[Fattal, 2008]: every patch has uniform reflectance, the appearance of the pixels within the patch expressed in terms of shading and transmission.
}shading and transmission signals unrelated;
} estimate the appearance of each patch by ICA;
}Failure when mag. of the surface reflectance is much smaller than that of the air light;
}[Tan, 2008]: Divide the image into a series of small patches and the corresponding patch in radiance should have a higher contrast;
}neighboring pixels should have similar transmission values formulated in a Markov Random Field, solved by graph-cut or belief propagation;
}Produce over enhanced images in practice.
}[He et al., 2009]: dark channel prior (easy to apply bright channel prior);
}The transmission of each patch estimated as the minimum color compo;
}Soft matting to ensure neighboring pixels had similar transmission values.
}[Kraz & Nishino, 2009]: use natural statistics of albedo and depth;
}Factorial MRF with statis. independent latent layers for albedo and depth;
}Solved a MAP problem by EM.
}[Tarel & Hautiere, 2009]: bilateral filter to factor radiance and airlight;
}Works not well at the depth discontinuities; could be better with guided filter?
}Results: Dehazing with dark channel prior and guided filter
Original
DeHazing
Original
DeHazing
Original
DeHazing
Original
DeHazing
Original DeHazing
Page updated
Google Sites
Report abuse