2013 Fourth International Conference on Emerging Intelligent Data and Web Technologies

A Combining Spatial Enhancement Method for Low Illumination Images

Wenjing Zhao

Zhong Cao

Pan Liu

Experiment Center, the Center of

School of Computer Science and

School of Energy Science and

Network and Modern Educational

Educational Software, Guangzhou

Engineering, University of Electronic

Technology, Guangzhou University

University

Science and Technology of China

Guangzhou, China

Guangzhou, China

Chengdu, China

[email protected]

[email protected]

[email protected]

Abstract-This paper proposes a combining spatial enhancement

method (CSEM) for low illumination images. The low
illumination images are low contrast and images grayscale
range are very narrow. There are two requirements of images
enhancement: enhancing image local contrast and adjusting
the dynamic range of the gray level. From the simulation
results, the proposed method shows better enhancement
performance and significant reduction in the computational
cost. So, the proposed method can be used in the images
captured by monitoring system in the low illumination
environment.
Keywords-spatial enhancement method; low illumination images;
image enhancement; images histogram; local contrast.

I. INTRODUCTION
Many monitoring and reconnaissance images are shot in
night without auxiliary light, so those images are usually
very dim and need enhancement processing. In order to
enhance the quality of low illumination images and visual
effects for human observing, or convert the images into
suiting machine-recognizable form, we need enhance the low
illumination images to obtain more useful information.
Image enhancement is to highlight some of the information
in the image according to certain requirements, and remove
or weaken those unwanted information.
A lot of image enhancement methods based on spatial
and frequency domain have been described in [1]. Reference
[2] proposed contrast enhancing method based on the
piecewise linear approximation of cumulative density
functions and reference [3] adopted a simple piecewise linear
function in order to sharpen the image edges and to reduce
the noise. A scheme for adaptive image contrast
enhancement based on a generalization of histogram
equalization has been proposed in [4]. A low-pass filter-type
mask has been used in [5] to get a nooverlapped sub-block
histogram equalization function to produce the high contrast
associated with local histogram equalization. A fast
multi-scale Retinex algorithm was presented in [6] to solve
the problem of color distortion and improve the
disadvantages of a slowly time-consuming arithmetic. Based
on the contrastive analysis on the RGB and the HSI color
space in [7], the transformation of Retinex algorithm
proposed to obtain a new effect on the color restoration.
In this paper, a combining spatial enhancement method
that can improve the quality of low illumination images is
described. We enhance low illumination images in two
978-0-7695-5044-2/13 $26.00 2013 IEEE
DOI 10.1109/EIDWT.2013.135

aspects: enhancing image local contrast and adjusting the

dynamic range of the image gray value.
II. LOW ILLUMINATION IMAGE ENHANCEMENT
The purpose of images enhancement is to emphasize
detail or characteristics of images. One images enhancement
method for all the images are effective, which is not present.
Firstly, it is a preliminary job for images enhancement to
understand the details and characteristics of the images. Then,
the theoretical basis of images enhancement are given, which
will be used. At last, according to the characteristics of the
images, we enhance the images to achieve the desired
targets.
ALow Illumination Image
The main purpose of the low illumination images
enhancement is to increase the contrast of the target and
background. The basic characteristics of the low illumination
image are the narrow dynamic range of the image gray value
accompanied by high random noise. The overall brightness
of image and the mean value of gray level are very low.
There are two low illumination images captured by
monitoring system in the low illumination environment and
those histograms are shown in Figure 1. From these figures,
we can see that the ranges of low illumination images
histogram are very narrow and below 50.

x 10

x 10

7
5

6
4

5
4

3
2

2
1

1
0

50

100

Figure 1.

743

150

200

250

300

50

100

150

200

Illumination grayscale image and histogram

250

300

 f  x, y    2 f  x , y 
L
0

(6)
g  x, y   
2
f
x
,
y


f
x
,
y
L

0






where L is the center coefficient of the Laplacian mask.
In fact, image enhancement is to use the following
Laplacian mask (Figure 2) to obtain as (6).

B Theoretical basis of the Enhancement

The mathematical thinking about the differential is
introduced to the image enhancement, and the strength of the
response of a derivative operator is proportional to the
degree of discontinuity of the image at the point at which the
operator is applied.
The differential in mathematics function has many forms.
However, we require that any definition we use for a
first-order differential (a) must be zero in flat segments
(areas of constant gray level values); (b) must be nonzero at
the onset of a gray level step or ramp; and (c) must be
nonzero along ramps. Any definition of a second-order
differential is similar to the first-order differential [1]. Since
this article is dealing with low illumination images whose
gray level are finite, the maximum possible gray level
changes also are finite, and the shortest distance over which
that change can occur is between adjacent pixels as in (1) as
shown. Robert conversion transform and Sobel transform are
the current Common gradient transforms based on the
first-order differential.
f
(1)
 f  x  1  f  x 
x
Similarly, the second-order differential is defined by the
following as in (2),
2 f
(2)
 f  x  1  f  x  1  2 f  x 
x 2
We know that second-order differential has pixel values
in more areas from (2). We can arrive at the conclusion:
Second-order differential has a stronger response to fine
details, such as thin lines and isolated points. This paper
considers the isotropic Laplacian operator as two-order
differential. For low illumination image enhancement
processing, the Laplacian transform of input image f  x, y 
is defined as follow:

2 f 

2 f 2 f

x 2 y 2

-1

-1

-1

-4

-1

(a) Laplacian mask

Figure 2.

(b) Another mask

Laplacian mask

Laplacian transform in the enhancing image detail, there

is a distinct advantage. In order to further highlight the
texture features of the low illumination images, we subtract a
blur version of an image from the image itself, which can be
expressed as the following (7):
(7)
f L  x, y   f  x, y   f  x, y 

where f L  x, y  is sharpening image by unsharp masking

f  x, y  is fuzzy form of f  x, y  .
A further generalization of unsharp masking is called
high-boost filtering. A high-boost filtered image, f hb  x, y  ,

is defined at any point

 x, y 

as shown in (8):

f hb  x, y   Af  x, y   f  x, y 

(8)

where A  1 , from (7), we obtain (9), which can be

expressed for computing a high-boost filtered image,

f hb  x, y    A  1 f  x, y   f L  x, y 

(3)

(9)

If the Laplacian transform is used, then we know that

f s  x, y  can be obtained using (6). In this case, (9)
becomes

In order to be useful for digital image processing that

equation (3) needs to be expressed in discrete form. Taking
into account that we now have two variables, we use the
following notation for the partial second-order differential in
the x -direction,
2 f
(4)
 f  x  1, y   f  x  1, y   2 f  x, y 
x 2
and, similarly in the y -direction, as

2 f
 f  x, y  1  f  x, y  1  2 f  x, y 
y 2

2

 Af  x, y    f  x, y 
g  x, y   
2

 Af  x, y    f  x, y 

L<0
L>0

(10)

where L is the center coefficient of the Laplacian mask.

High-boost filtering can be implemented with one pass
using either of the two masks as shown in Figure 3.

(5)

Because the Laplacian is a differential operator, its use

highlights gray level discontinuities in an image and
deemphasizes regions with slowly varying gray levels.
Background features can be recovered while still
preserving the enhancement effect of the Laplacian operation
simply by adding the original and Laplacian image. If the
center coefficient is negative, then we subtract, rather than
add, the Laplacian image to obtain an enhancement result.
Thus, the basic way in which we use the Laplacian for image
enhancement is as follows:

-1

-1

-1

4+A

-1

-2

-1

-1

(a) High-boost filtering

Figure 3.

744

bSobel gradient transform

High-boost filtering and Sobel gradient transform

CProposed Method
In order to enhance low illumination images, we can
process in following aspects: enhancing image local contrast
and adjusting the dynamic range of the gray level.
Considering the images enhancement effect and computing
speed, a combining spatial enhancement method is proposed
for low illumination images, that is high-boost filtering
method and the gradient transform method, which are based
on second-order differential and first-order differential,
respectively. The combining spatial enhancement method
(CSEM) is as follow:
Step 1. Firstly, we obtain the image 2 f  x, y  from

just as shown in Figure 4.

the image original grayscale image f  x, y  by high-boost

filtering, applying the high-boost filtering template (Figure
3(a), where A  1 ). Then, according to equation (6), we add
the image 2 f  x, y  to image f  x, y  and obtain the
image f L  x, y  , which has been sharpened.

Step 2. Image f s  x, y  is obtained from the original

image using the Sobel Gradient transform mask (Figure 3(b)).
Then, the smoothing image f s  x, y  obtained from the

Figure 4.

image f s  x, y  using a 4 *2 averaging filter.

III. SIMULATION RESULTS

We take the second low illumination image from Figure
1, which was captured by monitoring system in the low
illumination environment. Comparing the proposed
enhancement method with traditional low illumination image
enhancement methods, the results can be shown in Figure 5.
The simulation environment is MS windows 7, 64-bit and
Matlab2011b, 64-bit, simulation time is seconds.
The traditional low illumination images enhancement
algorithms are piecewise linear transform (PLT) and
multi-scale Retinex algorithm (MSR).

Step 3. Multiplying the smoothing image f s  x, y  to

the sharpening image

Flow chart of CSEM

f L  x, y  , we can obtained the

masking image GsL  x, y  .

Step 4. We find a threshold value T from the masking
image GsL  x, y  such that the number of pixel value more
than T but less than one percent to total of all pixels. Then,
we use (11) to obtain the enhanced image g  x, y  through
transforming the image to the gray scope of [0, 255]. If
decimal number is obtained in g  x, y  , that can be rounded
down.
0
GsL ( x, y)  0


 G  x, y  *255
(11)
g  x, y    s  L
0  GsL ( x, y) & &GsL ( x, y)
T
T

255
GsL ( x, y)  T

Remark:
(1)

(2)

We adopt Laplacian transformation in the

high-boost filtering, which obtains good results for
the image detail enhancement.
In the grayscale variation region, the image
GsL  x, y  is the product of image f L  x, y  and

image f s  x, y  , whose value is amplified.

Compared to the high-boost filter, the enhancement
to the noise and details in smoothing area by
smoothed Sobel gradient transform is finite. So, we
enhance image local contrast in step 3.
(3) We adjust the dynamic range of the grayscale from
narrow range to range of [0, 255] in step 4.
In summary, this paper offers the flow chart of
combining spatial enhancement method for low illumination

aOriginal gray image

bPLT

cMSR

dCSEM

Figure 5.

Contrast-enhanced low-illumination images

From Figure 5, we can find that three methods are effect

for low illumination enhancement. But our proposed method

745

is better than the others. In order to further illustrate, the

results are shown in Table I.

[7]

TABLE I. THREE METHODS COMPARING

EC
OI
PLT
MSR
CSEM

Clarity
0.63
1.58
1.88
6.10

IE
2.8
2.75
2.80
3.53

MV
12.03
30.12
36.12
53.19

SD
12.98
30.78
38.94
70.72

ET

0.105495
1.469766
0.239301

where EC is evaluation criterion, IE is information entropy,

MV is mean value, SD is standard deviation and ET is
elapsed time.

Figure 6.

histogram of Figure 1 enhanced using CSEM

We can find from Table 1 that greater enhancement has

been obtained by CSEM than PLT and MSR in clarity,
information entropy, mean value and standard deviation.
From Figure 6, the dynamic range of grayscale has been
adjusted to [0, 255].
IV. CONCLUSION
We have proposed a new image enhancement method
suited for low illumination images. It has been seen that the
new method can enhance the local contrast greatly and adjust
the dynamic range of the gray level, which are more
important to low illumination images. The proposed method
can be used in the images captured by monitoring system in
the low illumination environment.
REFERENCES
[1]

Rafael C. Gonzalez, Richard E. Woods, Digital Image Processing,

Second Edition, Pubishing House of Electronics Industry, Beijing,
2008

[2]

Sang-Yeon Kim, Dongil Han, Seung-Jong Choi and Jong-Seok Park,

Image Contrast Enhancement Based on the Piecewise-linear
Approximation of CDF, IEEE Transactions on Consumer Electronics,
vol. 45(3),pp.828-83, 1999

[3]

Fabrizio Russo, Piecewise Linear Model-Based Image Enhancement,

EURASIP Journal on Applied Signal Processing:12, pp.1861-1869,
2004

[4]

J. Alex Stark, Adaptive Image Contrast Enhancement Using

Generalizations of Histogram Equalization, IEEE Transaction on
image processing , vol. 9(5),pp. 889-896, 2000

[5]

Joung-Youn Kim, Lee-Sup Kim, and Seung-Ho Hwang, An

Advanced Contrast Enhancement Using Partially Overlapped
Sub-Block Histogram Equalization, IEEE Transaction on circuits and
systems for video technology, vol. 11(4), pp.475-484, 2001
Wen Wang, Bo Li, Jin Zheng, Shu Xian, Jing Wang, A Fast
Mutil-Scale Retinex Algorithm For Color Image Enhancement,
International Conference on Wavelet Analysis and Pattern
Recognition, pp.80-85, 2008

[6]

746

Jing Wu, Ziwu Wan g and Zhixia Fang, Application of Retinex in

Color Re storation of Image Enhancement to Night Image,
International Congress on Image and Signal Processing, pp.1-4, 2009