Post on 06-Jul-2020
Image enhancement based on selective-retinex fusion algorithm
Han-Sol Kang
ISL Lab Seminar
*Jin, Xuebo, Jia Bao, and Jingjing Du. "Image enhancement based on selective-retinex fusion algorithm." Journal of Software 7.6 (2012): 1187-1194.
*
INDEX
2019-04-10
01 Retinex Theory
02 Single-Scale Retinex (SSR) & Multiple-Scale Retinex (MSR)
03 Selective-Retinex Fusion Algorithm
04 Experimental Results
2
2019-04-10
01 Retinex Theory
3
What colors are this dress?
* Land, Edwin H., and John McCann. "Lightness and retinex theory." JOSA 61.1 (1971): 1-11.
*
But actually the answer is Blue and Black!!!
We can explain this problem using Retinex theory.
2019-04-10
01 Retinex Theory
4
What’s the meaning of Retinex Theory.
* Land, Edwin H., and John McCann. "Lightness and retinex theory." JOSA 61.1 (1971): 1-11.
*
Retinex = Retina + Cortex
Both the eye and the brain are involved in the Image processing.
A B=
If we just use the eye,
A B≠
But actually we use both the eye and the brain
2019-04-10
02 Single-Scale Retinex (SSR) & Multiple-Scale Retinex (MSR)
5
SSR
),(),(),( yxRyxLyxI
),(log),(log),(log yxLyxIyxR
)),(*),(log(),(log),(log yxIyxGyxIyxR
),(),(1
yxRwyxR ssr
N
n
nmsr
MSR
We can use the log function.
)),(),(log()),(log( yxRyxLyxI
2019-04-10
03 Selective-Retinex Fusion Algorithm
6
t
bxaf
2
)*arctan(5.0
S curve
[about b] [about a] [about t]
2019-04-10
03 Selective-Retinex Fusion Algorithm
7
t
bjiRajiPR
2
)),(*arctan(5.0),(
The modified Retinex algorithm
1) Light part need to reduce the brightness and its halo.
2) Dark part need to be enhanced according to the distance between the light source to keep distance information
BUT we can also see the shade information is eliminated!!!
So we need to recognize two different part.
[Source] [Modified Retinex]
2019-04-10
03 Selective-Retinex Fusion Algorithm
8
The selective and nonlinear gray mapping
[Four steps]
1) Find the light sources in the image.
erode first to eliminate speckles, then dilate to recover the area.
we can obtain the point light source 𝑃𝑛 , 𝑛 = 1…𝑁.
2) Reduce the halo.
compute the luminance-enhanced factor related to the distance.
NnM
jjiicjif
n
nn
T ...1,)()(
expmin),(
2
0
2
0
3) Deal with the two part differently.
compute the luminance-enhanced factor related to the luminance.
imagetheinpartsother1)),((
sourcelightpointeachofareathein1),(
2LightjipdjifL
otherwise3
0.15)thanlessisaverageluminancedark(theveryisimage6d
4) Enhance the luminance component of the whole image by using 𝑷𝑻(𝒊, 𝒋).
compute the luminance-enhanced factor related to the luminance.
),(),(),(),( jifjifjipjiP TLT
2019-04-10
03 Selective-Retinex Fusion Algorithm
9
The selective and nonlinear gray mapping
[luminance image of the source] [luminance image by the method]
2019-04-10
03 Selective-Retinex Fusion Algorithm
10
Selective-Retinex Fusion Algorithm
),()1(),(),( jiPgjiPgjiP TRY
otherwise6.0~3.0
distancenobservatiovisualandconditionlightgood3.0~1.0g
t
bjiRajiPR
2
)),(*arctan(5.0),(
),(),(),(),( jifjifjipjiP TLT
[Source] [Selective-Retinex Fusion Algorithm]
2019-04-10
04 Experimental Results
11
[Source] [Reference] [S-Retinex Fusion]
[1] method [2] method
[1] Songfeng Yin, Liangcai Cao, Yongshun Ling, Guofan Jin,“One color contrast enhanced infrared and visible image fusion method,” Infrared Physics & Technology, vol. 53, pp. 146–150, 2010.
[2] Alexander Toet. “Natural color mapping for multi and night vision imagery,” Information Fusion, vol. 4, pp. 155–166, August 2003.
2019-04-10
04 Experimental Results – modified retinex algorithm
12
[Source]
[SSR]
[HE] [BBHE]
[Gamma Correction]
gamma= 2.2 a=0.7, b=0, t=1.5
[Modified SSR]
2019-04-10
04 Experimental Results – modified retinex algorithm
13 a=0.7, b=0, t=1.5
[Source]
[SSR]
[HE]
[Modified SSR]
[BBHE]
[Gamma Correction]
gamma= 2.2
2019-04-10
04 Experimental Results – modified retinex algorithm
14 a=0.7, b=0, t=1.5
[Source]
[SSR]
[HE] [BBHE]
[Gamma Correction]
gamma= 2.2
[Modified SSR]