1. No category

NEAR-INFRARED GUIDED COLOR IMAGE DEHAZING
Chen Feng
Shaojie Zhuo
Xiaopeng Zhang
WHY NEAR-INFRARED (NIR) HELPS?
Mist
Fog
Sabine Süsstrunk
WORKFLOW
Visible image
Haze
Liang Shen
NIR image
Smoke
RGB – NIR
Image Pair
Airlight Color
Estimation
Image Dehazing
(Optimization)
Haze-free
Color Image
VIS
NIR
OUTPUT
RESULTS & COMPARISON
LIR
0.1um
IMAGE DEHAZING
INPUT
1um
10um
RGB
Particle Size
NIR
He [1]
Schaul [2]
Ours
NIR penetrates deeper into the atmosphere and preserves more details of distant objects.
CONTRIBUTIONS
• An optimization framework that resolves the
image de-hazing problem guided with NIR
gradient constraints.
• Better airlight color estimation by exploiting
the differences between NIR and RGB
channels.
PROBLEM FORMULATION
Air-light color estimation
Haze model
I p  t p J p  (1  t p ) A
Criteria for finding
local patch Ω:
A  arg min C ( J , t )
2
( x , y )
H  min{ min ( I ), D} D  N{| max ( I )  I
k
k
k{G , B}
k{ R ,G , B}
t1  Rc,n  t2
&
NIR
t3  I
NIR
 t4
&
|}
|| pt  pt0 ||  t5
2
Optimization framework:
RGB 2
NIR 

2
ˆ
ˆ
( J , t )  arg min || tJ  (1  t ) A  I || 1w | J  I | 2 | J | 3 || t ||
( J ,t )
ICIP 2013
Paper #2705
References
[1] K.M. He, J. Sun, and X.O. Tang, “Single image haze removal using dark channel prior,” IEEE Conference on Computer Vision and Pattern Recognition (CVPR), 2009.
[2] L. Schaul, C. Fredembach, and S. Süsstrunk, “Color image dehazing using the near-infrared,” Proc. IEEE International Conference on Image Processing (ICIP), 2009.
{chenf, szhuo, xiaopeng, liangs}@qti.qualcomm.com
sabine.sü[email protected]