Method and system for image haze removal based on hybrid dark channel prior

Patent metadata

Field	Value
Publication number	US-9361670-B2
Application number	US-201414476751-A
Country	US
Kind code	B2
Filing date	Sep 4, 2014
Priority date	Sep 4, 2014
Publication date	Jun 7, 2016
Grant date	Jun 7, 2016

Abstract

A method and a system for image haze removal are provided. The method includes the following steps: receiving an input hazy image including input pixels; determining a hybrid dark channel for each of the input pixels according to a first minimum dark channel of a corresponding large local patch and a second minimum dark channel of a corresponding small local patch; determining a transmission map for each of the input pixels according to the hybrid dark channel prior, the corresponding hybrid dark channel, and atmospheric light associated with the input hazy image in each color channel; determining a color spectrum adjustment parameter corresponding to each of the color channels; recovering scene radiance for each of the input pixels in each of the color channels according to the corresponding color spectrum adjustment parameter, the transmission map, and the corresponding atmospheric light to produce and output a de-hazed image.

First claim

What is claimed is:

1. A method for image haze removal based on a hybrid dark channel prior, adapted to an electronic device, comprising:
receiving an input hazy image comprising a plurality of input pixels;
determining a hybrid dark channel for each of the input pixels according to a first minimum dark channel of a corresponding large local patch and a second minimum dark channel of a corresponding small local patch;
determining a transmission map for each of the input pixels according to the hybrid dark channel prior, the corresponding hybrid dark channel and atmospheric light associated with the input hazy image in each color channel, comprising:
for each of the input pixels:
calculating the first minimum dark channel by performing a minimum operation on each large local patch pixel of the corresponding large local patch in each of the color channels and performing a minimum filter on the corresponding large local patch;
calculating the second minimum dark channel by performing the minimum operation on each small local patch pixel of the corresponding small local patch in each of the color channels and performing the minimum filter on the corresponding small local patch;
allocating a first weight to the first minimum dark channel and a second weight to the second minimum dark channel so as to obtain a first weighted minimum dark channel and a second weighted minimum dark channel, wherein the first weight and the second weight are positive values between 0 and 1, and a summation of the first weight and the second weight is 1; and
combining the first weighted minimum dark channel and the second weighted so as to obtain the hybrid dark channel;

determining a color spectrum adjustment parameter corresponding to each of the color channels;
recovering scene radiance for each of the input pixels in each of the color channels according to the corresponding color spectrum adjustment parameter, the transmission map, and the corresponding atmospheric light so as to produce a de-hazed image; and
outputting the de-hazed image.

2. The method according to claim 1 , wherein the formula for determining the hybrid dark channel for each of the input pixels according to the first minimum dark channel of the corresponding large local patch and the second minimum dark channel of the corresponding small local patch comprises Eq.(1):

J
dark

⁡

(

x
,
y

)

=

α

α
+
β

⁢

min

(

i
,
j

)

∈

Ω
⁡

(

x
,
y

)

⁢

(

min

c
∈

(

r
,
g
,
b

)

⁢

J
c

⁡

(

x
,
y

)

)

+

β

α
+
β

⁢

min

(

i
,
j

)

∈

μ
⁡

(

x
,
y

)

⁢

(

min

c
∈

(

r
,
g
,
b

)

⁢

Assignees

• Univ Nat Taipei Technology

Inventors

• Huang Shih-Chia
• Chen Bo-Hao
• Cheng Yi-Jui

Classifications

• G06T2207/10024

  Color image · CPC title

• G06K9/4652

  Physics · mapped topic

• G06T2207/30232

  Surveillance · CPC title

• G06T2207/20024

  Filtering details · CPC title

• G06T5/009

  Physics · mapped topic