2011 The 6th International Forum on Strategic Technology

Industrial X-Ray Image Enhancement Algorithm

based on Adaptive Histogram and Wavelet
L J' W Y W ' W L ' 1 Il, ang an el' , ang el
l. Automation College, Harbin Engineering University,
2. Huarui College, Northeast Petroleum University
Harbin, China
Abstract-In this paper, we an industrial X-ray image enhance
algorithm based on histogram and wavelet is proposed in this
paper. The proposed technique is able to cope with low contrast
and poor feature details. Firstly, the original image is divided
patched by the contrast limited adaptive histogram equalization
algorithm in order to adjust whole contrast. Then a map is built
between the image and the detail scales by the wavelet ratio in
order to adjusting the local contrast. Finally, according the
enhance function radios, the reconstruct image is rebuild.
Compared with other image enhancement algorithms,
experimental results show that our algorithm can improve the
global image contrast effectively as well as overcome the these
visible artifacts of X-ray image, the x-ray image become more
clear, and a better perceptual image is acquired for the image
feature recognize and matching.
Keywords- X-ray; Wavelet; Adaptive; Image Enhancement
1. INTRODUCTION
The valid method to enhance X-ray image is histogram
enhancement. The histogram of industry X-ray in a small fled,
most of pixels in the low fequency areas, and there are hardly
few pixels in high fequency areas, Even the industry X-ray
image is an image seems like the black image. Therefore, it is
need a method to image enhancement. Recently, many people
researcher on this feld. Shu Yang and Cairong Wang proposed
an image enhancement algorithm based on multi-scale
morhological reconstuction [1]. STEPHANE O. MLLAT
researches the theory for multi-resolution signal decomposition
[2]. Pizer S M proposed adaptive histogram equalization and its
variations [3]. Tang Jinshan adopted a direct image contast
enhancement algorithm in the wavelet domain for screening
mammograms [4]. Wang Xiu-bi refers to image enhancement
based on lifing wavelet transfor [6]. Jinshan Tang and
Qingling Sun used contast measure in the wavelet domain for
screening mammograms image enhancement algoritm. Tn this
paper, we will investigate image-enhancement technology
based on adaptive histogram and wavelet.
II. LIMITED CONTRAST ADAPTIVE HISTOGRAM
ENHANCEMENTALGOmTHM
A. Hstogram Enhancement Theory
Histogram enhancement, the advantages of high speed and
better effect, wildly applied in X-Ray processing. Histogram is
the fnction of gray levels, which denotes the gray level of
978-1-4577-0399-7111/$26.00 2011lEEE 836
Wei Jie , Wang Da Da
3. Yunan power test institute (group) Ltd,
Kunming, China,
every pixel. Therefore, the contrast radio will be improved by
gray nonlinear tansform to adjust the accumulation fnction,
and the gray in small range will be tansfor in the whole fled.
Image histogram is an important tool to analysis image gray
distribution, which is a summary graph showing a count of the
data points failing in the data. Therefore, histogram is defmed
as follows.
(1)
Which n is the total pixels of X-ray image, _ is the ,_ gray
level of coresponding pixels.
The X-ray is the low contast, which most pixels are in the
same gray level, thus it is hard to recognize. Histogram
equalization is a method to improve image contast it built a
nonlinear tansfer fnction to tansform the histogram of
original image. The processed image with equal gray scale in
the same gray level, at the same time, the maximum entopy is
acquire, it means that the image information is abundant.
Hypothesis that the gray tansfer fnction is s = fer) ,
which slope is the limited non-minus continuum decline
fnction, it can tansf input image A( y) to output image
B(x,y), the HA(r) is the histogram of input image, and
H
1
(r) is the histogram of outut image. According the
defnition of the histogram, the original image histogram and
processed histogram areas are equal:
H
1
(s)ds = H A (r)dr (2)
The relationship between H A (r) and H
1
(r) is the following:
,
R
Where r (r) = HA (r) is constant.

A
Afer transfored, every pixel of the image is tensile to
extend the image gray level areas, and the vision effect is
improved.
August 22-24,2011
B. Adaptive Histogram Enhancement Algorithm
Compared with the whole histogram equalization, the
adaptive histogram enhancement (AHE) has the advantage of
good local contrast. But the ARE algoritm will commute the
local histogram and accumulation distributing fnction of
every pixels, it is extremely computational intensive. Besides,
the AHE algorithm is sensitive for noises, which is liable to
bad enhancement for local area.
4
g
(x,
Y
)
= z7
(
r
(x,y))
(4)
i=
l
T(r(x,y)) is the gray transfor fnction, r(x,y) is the gray
value of pixel (x,Y) .this image enhancement algorithm based
on histogram, independent with the pixel position.
TIT. `-1Y^^OIlH^lCI^Il11IOCKI1H^H^5IICl
L Adaptive Histogram of Limited Contrast Algorithm
1I^!1INII51COK^^^lIY^NIII1
AR algorithm improve the image contast, at the same
time magnifed the noise. Even the enhancement will lead to
image distortion in some detail area, which lead to noise
amplifcation and distortion, that affects the image diagnose.
The frst step of AHE algorithm is input image A(x,y) block.
and the local histogram in every block. Thus limit contast
fnction to AR in every block to generate tansform fnction
respectively. Then the adaptive bilinear interolation used to
joint output imageB(x,y).
It is need to preliminary adjustent before the image block,
Defmite the limit fnction to limit the gray level probability
density, and adjust the exceed histogram, fg.! shows the
processing.
pixel
pixel
gray level
___ --
gray level
(a)original gray chat (b )adjusted gray chat
Fig.l limited function
To decrease the block effect, it is need to bilinear
interolation. The block gray value is the gray transfor
fnction of adjacent pixels,
( x,y)
WI
W2
W4 W3
F ig.2 adapative interpolation processing
Fig.2 depicts the bilinear interolation processing, (x, y) is
the center of the sub image, compute the areas scale in adjacent
areas, the value is the following:
J7
Wavelet tansfor based on fourier tansfor, appearance
in 80thof 20 cenry, with such better characters in time and
fequency felds, it is widely applied in image
processing ,model identif and robot vision and so on. In 1989,
Mallat promote the tower algorithm [7], which can disassemble
the image into some sub images according to different scale,
orientation and space feld. In particular, this material is used in
this book for image data compression and for pyramidal
representation, in which images are subdivided successively
into smaller regions.
A. Relationshi between the image and the wavelet
Image signal is the two-dimension signal, therefore
secondary wavelet to analysis multiple resolution is used in this
paper. Because of the image is disperse signal, the wavelet
transfor is the two dimension wavelet disperse tansform,
which is derived fom the one-dimension wavelet transform.
The principle of wavelet tansfor is multiple resolute
decomposing. The image is divided into some sub images in
different space and fequency, then quantization and coding.
The total of sub images data is equal to the original image data,
but the statistical characteristics of them are different. In this
processing, the key point is the best optimum wavelet base.
B. Selection and Optimization of Wavelet Base
Wavelet base collection is important for image compressing.
Image is different fom other images in quality, any losing
details will lead mistake diagnose. Wavelet tansfors can
comeback original image signal, but different wavelet bases
have different results, the compressed ratio is different. The
algorithm of wavelet base collection is not the unique, but
every wavelet base has it own characteristic.
If (x) is a one-dimension scale fnction, I(x) is
coresponding wavelet quotiety. Then the two-dimension
wavelet transform fnction is like eq 1 to eq4.
I
l
(x, y)
=
(x)I(Y)
1
2
(x, y)
=
I(x)(y)
I
\
x,y)
= l
(x)
l
(y)
(x,y)
=
(x)(y)
(5)
(6)
(7)
(8)
At the two dimension conditions, there is a scale
fnction (x,y) , which fex and tanslation will become a
group of orthogonal base. Assumption that
2
i (x) is the
orthogonal base of the set
(r)
2, (x-r) n, y- r
}
m)) ,
(n, m)E Z20f the V2
)
'
For a cluster of detachable multi-resolution approximation
on L2 (R x R) , every vector space disassemble to the same
two subspace tensor product of L2 ( R) , while the V2i
compose the multi-resolution approximation of L2 (R) , the
vector sequence V2
}
compose the multi-resolution
approximation of L2 (RX R) . That is
(x
, y) =
(x
)
(y) is
true. Among this *
(x) is the coresponding one-dimension
scale fnction. Then the fnction is
!(x
,y) , among the 2j
resolution approximation, as follows.
The resolution is 1, which N is global pixel. Obviously, the
disperse image approximation have 2j pixel, liking the one
dimension circumstance, hypothesis the level and the
perendicularity is symmetry of the image. Afer 2j resolution,
the detail signal is the projection of orthocomplement space
Q2j of !(x,Y) of v2 , base.
The information difference between the A
f
, +
d
and A
f
,
!
IS equal to projection Q2f of !(x,y) , which can express by
inner product as following.
DL+
d
=
!(x,Y),I
j
(x-r
}
n,y-r' m)
(10)
Di
} +
d
=
!(x,Y),I
J
(x-r) n,y-r
}
m) (11)
D, +
d
=
!(x,Y),I
j
(x-r' n,y-r' m) (12)
Therefore, two dimension approach the Af
,
! , and the detail
is the convolution of
D1+1
!
' Di+l!
' D+l!
'
} } }
DJ=f(x,Y)*I2j( -x)I2/-Y)2k2 in,-2 1m)
(16)
Where (n,mE Z
2
) , From the EQ.9 to EQ.l2, the two
dimension image can disassemble the main part of Af! and
,
the details of DL+l!
' DL+
d
and
D
L+d, which the
details can acquire by fltering along the x direction, y direction
and diagonal orientation, thus fltering disassemble is a group
of signals disassemble in fequency direction. Which resolution
fequency is lowest in 2j and AfJ' and then given the x and
y direction low fequency, while
DL+l!
' DL+l! and
838
D
'
+l! given the level, perendicularity and the diagonal
direction high fequency separately. Finally, the image Af!
}
can disassemble the 3J+ 1 dispersed sub images.
(AfJ,/(DL!) - J<j<J,/(Di
j
!) - J<j<J,/(D
j
!) - J<j<)
(17)
This is the processing of wavelet biorthogonal disassemble,
which 21 is sketchy approximation,
DL+
d
,
Di}+
d
and
D} +d is the detail signal in the same resolution of different
directions. Tn this algorithm, the total pixel is invariability.
Fig.3 Algorithm fow chart
Fig.3 is the algorithm fow chart of image enhancement
based on adaptive histogram and wavelet tansfor. The main
idea is that founded a limited fnction to adjust the image
contast , then record the position and the gray value, then
wavelet transfor is used to adjust the enhancement ratio.
IV. EXPERIMENTS AND CONCLUSION
Fig4 (a) is the original input image, which IS the
noralized industial X-ray Image, which sIze IS
256*256.From the image.
(a) input image (b ).histogram
Fig.4lnput image of X-ray
From the FigA, we can see that the contast is very low
that it is hardly to see anything in the FigA (a). Fig.5 is the
wavelet transfored results.
(a) input image (b ).histogram
Fig.5 Wavelet enhacement transform
The processing time is 1.03600 I s. The enhancement is
poor, the processing image also with low contrast, and hardly
to recognize.
(a) input image (b ).histogram
Fig .6 r adjust enhancement
Fig.6 is the r adjust enhancement results. The processing
time is 1.705392s, and the image contrast is enhanced slightly.
Fig6 (b )is the histogram of r adjust enhancement algorithm,
we see that the gray feld is enlarged in a small feld.
(b ).histogra
Fig.7 AHE algorithm enhancement
Fig7 is the ALE algorithm results. The processing time is
8.676029s. The image contrast is enhanced. Fig7 (b) is the
histogram of ALE algorithm. From the Fig7, we see that the
gray feld is enlarged in different feld. But the back ground of
the fg7 is complied mixed with noises.
1
(a) input image (b ).histogra
Fig.8 Enhancement based on Adaptive Histogram and Wavelet
839
Fig8 is the algorithm results. The elapse time 4.68264 s. the
image contrast is enhanced. Fig8 (b) is the histogram of ALE
algorithm results. We see that the gray feld is enlarged. From
the results, we can conclusion that our algoritm can improved
the global contast, at the same time, it is obviously that image
enhanced between the detected element and background.
V. CONCLUTIONS
Industial X-ray widely applied in nondestructive
testing, Industial X-ray image has the characteristics of low
contast and more details, thus it is need to an image
enhancement algorithm. To enhance the image contast, limited
adaptive histogram equalization algorithm is used in the frst
step. We set the initial limit coeffcient of limit fnction then
building a map between the image and the detail scales by the
wavelet ratio, which is adjusted by the local contracting
information. According the enhance fnction radios, the
reconstruct image is rebuild. Compared with other image
enhancement algorithms, experimental results show that our
algorithm can improve the global image contrast effectively,
moreover, restrain the background and enhance the contrast
between the background and the detection element. Processed
image, with more details inforation and better vision, benefts
to frther identit and recognize.
ACKNOWLEDGMENT
The paper is supported by the Y unan power institute of
the Yunan power test institute (group) Ltd.
REFERENCES
[I] Shu Yang .Cairong Wang . Ligou Deng .A new Approach of Image
Enhancement Based on Multi-scale Morphological Reconstruction.
Hybrid Intelligent Systems. HIS09. Ninth International Conference
on.pp.II3-116
[2] STEPHANE G. MALLAT.A Theory for Multiresolution Signal
Decomposition:The Wavelet Representation.IEEE TRANSACTIONS
ON PATTERN ANALYSIS AND MACHINE INTELLIGENCE. VOL
I I . NO. 7. pp. 674-693,JUL Y 1989
[3] Pizer S M. Ambtrrn E P. Adaptive histogram equalization and its
variations.[J].Computer Vision Graphics & Image
Processing ,1987,39(3):355-368.
[4] Jinshan Tang; Xiaoming Liu; Qingling Sun; A Direct Image Contrast
Enhancement Algorithm in the Wavelet Domain for Screening
Mammograms. Selected Topics in Signal Processing, IEEE Journal of
Volume: 3 , Issue: I ,pp. I
[5] Xiu-bi Wang. Image enhancement based on lifing wavelet transform.
Computer Science & Education, 2009. ICCSE 09. 4th International
Conference on : 2009 , pp. 739 - 741
[6] linshan Tang, Qingling Sun, Kwabena Agyepong. AN IMAGE
ENHANCEMENT ALGORITHM BASED ON A CONTRAST
MEASURE IN THE WAVELET DOMAIN FOR SCREENING
MAMMOGRAMS. Image Processing, 2007. ICIP 2007. IEEE
International Conference on. 2007 , pp. 74-80
[7] Tsai R T,Huang T S. Multifame image restoration and
registration[J].Advaces in Computer Vision and Image
Processing,1984(1):317-319