MODIFIED ALPHA-ROOTING COLOR IMAGE ENHANCEMENT METHOD ON THE TWO-SIDE 2-DQUATERNION DISCRETE FOURIER TRANSFORM AND THE 2-DDISCRETE FOURIER TRANSFORM

Applied Mathematics and Sciences: An International Journal (MathSJ ), Vol. 4, No. 1/2, June 2017
DOI : 10.5121/mathsj.2017.4201 1
MODIFIED ALPHA-ROOTING COLOR IMAGE
ENHANCEMENT METHOD ON THE TWO-SIDE
2-DQUATERNION DISCRETE FOURIER TRANSFORM
AND THE 2-DDISCRETE FOURIER TRANSFORM
Artyom M. Grigoryan 1
, Aparna John 1
, Sos S. Agaian 2
1
University of Texas at San Antonio, San Antonio,TX 78249, USA
2
City University of New York / CSI
ABSTRACT
Color in an image is resolved to 3 or 4 color components and 2-Dimages of these components are stored in
separate channels. Most of the color image enhancement algorithms are applied channel-by-channel on
each image. But such a system of color image processing is not processing the original color. When a color
image is represented as a quaternion image, processing is done in original colors. This paper proposes an
implementation of the quaternion approach of enhancement algorithm for enhancing color images and is
referred as the modified alpha-rooting by the two-dimensional quaternion discrete Fourier transform (2-D
QDFT). Enhancement results of this proposed method are compared with the channel-by-channel image
enhancement by the 2-D DFT. Enhancements in color images are quantitatively measured by the color
enhancement measure estimation (CEME), which allows for selecting optimum parameters for processing
by thegenetic algorithm. Enhancement of color images by the quaternion based method allows for
obtaining images which are closer to the genuine representation of the real original color.
KEYWORDS
Modified alpha-rooting, quaternion Fourier transform, color image enhancement
1. INTRODUCTION
All color images consist of 3 or 4 channels of monochromatic images. The color of the image is
resolved to color components, and pixel values in each channel are the corresponding color image
intensities. For example in the RGB color model, the image is stored as three monochromatic
images for the red, green, and blue colors. The original color in the image is the color obtained by
superposing or merging all three color components. The processing of color images could only be
done with fairness if the color of the image is considered as the sum of the color component pixel
values in all channels when taken together. But until recently, the processing of the color image
was done by splitting up the color image channel-by-channel to individual images and, then,
processing each channel with the respective algorithm. To know the totality of enhancement
effects of the respective algorithm onto the color image, each enhanced channel is taken together
to compose the resulting enhanced image. Ell’s work of quaternion approach [11]
of color image
processing enables us to represent the color of the image as a sum of the color components.
Quaternion numbers are four-dimensional hyper-complex numbers. Each of the resolved color
components in the pixel can be represented as the component vector in four-dimensional space.
That is, a color image is represented as a quaternion image. This representation helps in
processing color image by taking the color of the image as the vector sum of color components.
The color of the image is considered as a single entity, rather than considering it as three separate
resolved color components. In this quaternion approach of color image processing, all three or

2
four channels in the color image are processed simultaneously. This type of enhancement method
gives a merging of all of the color components.
Many frequency domain based enhancement algorithms, particularly the Fourier transform-based
enhancement algorithms,[1]-[3], [26]-[29]
are effective on grayscale images and on color images, when
processing channel-by-channel. But there is a need to have efficient color image enhancement
algorithms, in which the color is taken as a single entity, or in other words when the processing is
done on all channels simultaneously. Many quaternion transform based enhancement methods,
which are based on the QDFT, give effective enhancement results. Since quaternions are four-
dimensional numbers, the QDFT are transforms in four-dimensional space. Therefore, in orderto
implement an enhancement algorithm with the QDFT, there is a need to Figure out fast
algorithms. The proposed fast algorithm[2,3,19,30]
of QDFT reduces the computational complexities
in enhancement algorithms based on the QDFT. The Fourier transform-based enhancement
algorithms, such as the alpha-rooting, log alpha-rooting, and modified alpha-rooting
methods,which are proven to be effective on grayscale images, can be extended to the quaternion-
based color image enhancement. The proposed algorithm is the modified alpha-rooting method
followed by different spatial transformation techniques, and the preliminary experimental results
show that the proposed enhancement algorithm is effective.
Choice of variables needed for the enhancement algorithm could be made easier by using an
optimization algorithm. In order to use such an algorithm, we need to get a cost function that
would give a quantitative measure of enhancement in the processed image. Grigory an and
Agaian proposed a quantitative measure for enhancements in the enhanced images after
processing with the respective enhancement algorithm. The quantitative measure is referred to as
the color enhancement measure estimation (CEME). The CEME value is used as the cost function
in the proposed optimization algorithm, which is chosen to be the genetic algorithm. This paper
describes the method and experimental results of the enhancement by the modified alpha-rooting
methods by the 2D-QDFT on color images. Color image processing is done by the quaternion
approach, in which all channels in the image are processed simultaneously. The experimental
results obtained by the quaternion approach are compared with results obtained by channel-by-
channel processing, by the same method of modified alpha-rooting, but with the traditional 2-D
DFT. Enhancement in images is measured quantitatively by the CEME value and the
optimization of variables is done with the genetic algorithm.
2. METHODOLOGY OF 2-DQDFT AND 2-DDFT MODIFIED ALPHA-
ROOTING
2.1. Quaternion Numbers
Quaternion numbers are four-dimensional hyper-complex numbers[9][10]
. A quaternion number
is represented in Cartesian form as
= + + + . (1)
That is, every quaternion number has a scalar part, ( ) = and a vector part, ( ) = + +
. The imaginary units i, j, k are related as
= = = = −1;
= − = ; = − = ; = − = .
(2)
The basis {1, , , } is the most common and classical basis to express a quaternion. There are
many different choices for . Given two pure unit quaternions µ and ψ, that is, μ = ψ = −1,

3
which are orthogonal to each other, that is, µ ┴ ψ, the set {1, µ, ψ, µψ} is a basis for any
quaternion .
The norm of the quaternion defined as
‖ ‖ = + + + . (3)
When ‖ ‖ = 1, q is said to be unit quaternion. When the scalar part, ( ) = 0, then the
quaternion is referred as a pure quaternion.The polar form of a quaternion number is given as
= | | = | |! "# $% + &% # ' $%(, (4)
where| | is the modulus and &% is called the axis and $% is the phase or argument of that are
expressed as
| | = )‖ ‖ = )( + + + ), &% =
* + + + ,
)( + + )
,
$% = tan01
2
)( + + )
3 .
(5)
An important property in quaternion algebra is the product of two quaternions, whichis not
always commutative. That is, for a given two quaternion numbers 4 and .
4 = 4 or4 ≠ 4. (6)
2.2.Color Image In The Quaternion Space
The color image can be represented as a quaternion image [9],11]-[16]
. Depending on the color
model, the color images have three or four channels. In the case of three channel color models
like the RGB, XYZ, or in luminance-chrominance color model YCbCr, the color images can be
represented as a pure quaternion. For example, the color image f (n,m) in RGB color model can
be represented as a quaternion image
8(', 9) = :(', 9) + ;(', 9) + (', 9), (7)
where (', 9), ;(', 9),and (', 9) are the red, green and blue components, respectively. In
general, for three or four channel color images, the parts , , , and of the quaternion number
take the form[56]
= 8<,=>?@<<AB 1
, = 8<,=>?@<<AB
, = 8<,=>?@<<AB C
, = 8<,=>?@<<AB D
. (8)
When a color image is represented as quaternion, we get a correlation between all color
components. In the usual color image enhancements, this correlation between the color
components is not used, because each channel is processed separately. The even component, E of
the quaternion color image can be chosen in many different ways. The two common choices are
= 0and the grayscale value, =
F
G
[: + ; + ].

4
2.3 Two-Side 2-Dquaternion Fourier Transform
The discrete Fourier transform of 2-D hyper-complex numbers is referred to as the 2-D QDFT.
As mentioned before, the product of two quaternions are not always commutative. Thus, based on
the position of the kernel of the transform with respect to the image, there are many versions of
2D QDFT. In the two-side 2D-QDFT, the image is sandwiched by the two transform kernels and
defined as
JKLM, L(4, #) = N OP
<Q
R N 8<,=OS
=T
U01
=VW
X
Y01
<VW
, 4 = 0: ([ − 1), # = 0: ( − 1).
(9)
The inverse transform is defined as,
]JKLM, 8(', 9) =
1
[
N OP
0<Q
RN LQ,TOS
0=T
U01
TVW
X
Y01
QVW
,
' = 0: ([ − 1), 9 = 0: ( − 1).
(10)
2.4 Modified Alpha-Rooting
In the alpha-rooting method of image enhancement [4] [17] –[20] [29] –[34]
, for each frequency point
(4, #), the magnitude of the discrete quaternion transform are transformed as
LQ,T → [LQ,T]_ (11)
There are many different options for the choice of the M and ` lies between (0,1) [4],[17]-[20],[34]
. The
modified alpha rooting method is an extension of the alpha-rooting with M as a function of
magnitude of the image defined as
= a(4, #) = b"; c
d|e(4, #)|f
+ 1g, 0 ≤ i, 0 < k. (12)
2.5 Color Enhancement Measure Estimation (Ceme)
The color enhancement measure estimation (CEME), is an enhancement measure [4]-[8]
based on
the contrast of the images. The proposed enhancement measure is related to the Weber law, which
basically explains that the visual perception of the contrast is independent of luminance and low
spatial frequency. To calculate the CEME value,a discrete image of size [ × is divided[21]
by
1 blocks of size m1 × m blocks each, where < = n[</m<p, for ' = 1,2. For the RGB color
model, when an image is transformed by the 2-D QDFT
f = (fs, ft, fu) → fv = !fw
x, fs
x, ft
x, fu
x(, (13)
where 8A
x, referring to the scalar component of the quaternion image, which is obtained after using
the transform, the CEME value is calculated by
y%(`) = ayy_!8v( =
1
1
N N 20 log1W
S|
BV1
S}
SV1
~
9 •S,B (8A
x, 8€
x , 8•
x , 8‚
ƒ)
9 'S,B (8A
x, 8€
x , 8•
x , 8‚
ƒ)
„.
(14)

5
Methodology
The images are first enhanced by the modified alpha-rooting method. We sawin the channel-by-
channel histograms of these enhanced images, that all intensity values of each of the channelwere
shifted to darker regions.To increase the brightness of these enhanced images, that is, to shift the
histograms back to the brighter side, the logarithmic or power-law (gamma) transformations are
done. The general case of logarithmic transformations[1]
is defined as
# = [log(1 + :)]Q
, (15)
where: is the input image and # is the resulting output image of the logarithmic transformation
and is a scaling factor. The simplest case is when 4 = 1.The power law (or gamma)
transformation[1]
is defined as
# = (:…), (16)
Where r is the input image and s is the resulting output image of the gamma transformation and c
is a scaling factor. When the input image itself is dark, all enhancement algorithms are performed
on the negative of the image. The positive of the resultant image is taken at the final stage of
image processing.
While processing with quaternion image, it is sometimes easier to see the quaternion image in the
polar form. From Eq. (4), the polar form of quaternion image is = | | . The spatial
transformations, such as the log transformation and the gamma transformation, modify the
magnitude without altering the phase of the quaternion image. If †<Q‡ˆ and ‰‡ˆQ‡ˆ are the
quaternion image before and after applying the spatial transformation techniques, such as the log
and gamma transformation techniques, then the transformed quaternion image is obtained by
‰‡ˆQ‡ˆ = Š ‰‡ˆQ‡ˆŠ
†<Q‡ˆ
Š †<Q‡ˆŠ
(17)
whereŠ †<Q‡ˆŠ and Š ‰‡ˆQ‡ˆŠ are the magnitude of the quaternion image before and after applying
the transformation techniques.
2.6. The Enhancement Measure Estimation (EME) For Grayscale Images
The processing of the image is done on each channel separately. The 2-D DFT and the
subsequently the modified alpha-rooting is done on each channel individually. The enhancement
of the image by the modified alpha-rooting causes the intensity of the histogram to be shifted
toward darker regions. But, further enhancement of the image by gamma transformation or
logarithmic transformation results in the intensities of the histogram that are shifted towards
brighter regions.
The enhancement measure estimation (EME) for grayscale images is an enhancement measure [4]-
[8]
based on the contrast of images. The idea behind this estimation measure is similar to that of
the CEME measure for color images, but in the EME only, one color channel or grayscale image
is considered. To calculate the EME value, the 2-D image of size [ × is divided[21]
by 1
blocks of size m1 × m each.Here, < = n[</m<p, for ' = 1,2. When an image is transformed by
the 2-D DFT,
8 → 8,x (18)
where 8v, referring to the enhanced image, the EME value is calculated by

6
y%(`) = yy_!8v( =
1
1
N N 20 b";1W
S|
BV1
S}
SV1
~
9 •S,B (8v)
9 'S,B (8v)
„.
(19)
3. EXPERIMENTAL RESULTS
3.1. Using 2-DQDFT
Figure 1is the enhanced image results of “A Streamlined Form in Lethe Vallis, Mars-
esp_045833_1845[35]
” was enhanced by the modified alpha rooting method. First the variation Of
the CEME for `, i, and λ values were plottedwith two parameters as variables and a fixed value
for the third parameter. Figure 1a shows the surface plot of the CEME values for different values
of `, and i, for the given λ=0.58. For the image “A Streamlined Form in Lethe Vallis, Mars -
esp_045833_1845”, the maximum value of the CEME for the modified alpha-rooting method of
enhancement is obtained at ` = 0.96, i = 0.93 for λ = 0.58.
We understood from Figure 1e that the intensity values of the histogram of the image enhanced
by modified alpha-rooting method were shifted to darker regions, as compared to the histogram of
the original image in Figure 1c. The CEME value obtained is higher, showing a better
enhancement in the contrast. To increase the visual perception of the image, the image obtained
from the modified alpha-rooting is further brightened by the logarithmic transformation. The
logarithmic transformation adopted for the enhancement of the image is shown in Figure 1d,
as# = [log 1 : IQ
, with 1, and 4 3.3. By doing the logarithmic transformation, the
histogram of the modified alpha-rooting image is shifted to brighter intensity regions.
Figure1:(a) Surface plot of CEME vs `, i for λ=0.58 for the image “A Streamlined Form in Lethe Vallis,
Mars- esp_045833_1845[35]
” –“
http://www.nasa.gov/mission_pages/mars/images/index.html (b) The
original image “A Streamlined Form in Lethe Vallis, Mars - esp_045833_1845*
”; (c) The histogram of the
original image;

Figure1:(continued)(d) The enhanced image by the method of 2
histogram of the image is (d). (f) The i
and
The logarithmic transformation of the image enhanced by the modified alpha
compared with the logarithmic transformation of the original image.
algorithms were also compared side
Figure 2: (a) Original Image (“A Streamlined Form in Lethe Vallis, Mars
*
http://www.nasa
Figure 2: (continued) (b) Image enhanced by the modified alpha
transformation; (c) The color image enhancement by Retinex algorithm (McCann
transformation of the original i
(d) The enhanced image by the method of 2-D QDFT Modified Alpha-Rooting; (e) The
The image enhanced by the logarithmic transformation of the image in (d)
and (g) the histogram of the image in (f).
The logarithmic transformation of the image enhanced by the modified alpha-rooting is
compared with the logarithmic transformation of the original image. Results of a few other
side-by-side are shown in Figure 2(a) to (e).
(a)
Figure 2: (a) Original Image (“A Streamlined Form in Lethe Vallis, Mars - esp_045833_1845
http://www.nasa.gov/mission_pages/mars/images/index.html);
(b) Image enhanced by the modified alpha-rooting followed by the logarithmic
transformation; (c) The color image enhancement by Retinex algorithm (McCann-99).(d) The logarithmic
mation of the original image; (e) The histogram equalization image of the original image.
7
Rooting; (e) The
of the image in (d)
rooting is also
of a few other
esp_045833_1845[35]
”
rooting followed by the logarithmic
(d) The logarithmic
image of the original image.

8
It could be inferred from Figures 2a to 2d that the image enhancement by the combined effect of
modified alpha rooting followed by logarithmic transformation (Figure 2b) gives a better visual
perception than the original image (Figure 2a), or the enhancement by the retinex algorithm
(Figure 2c) or enhancement by just the logarithmic transformation (Figure 2d). The enhanced
image by modified alpha-rooting combined with logarithmic transformation, could also reveal
some hidden details of the dark regions of the original image.
In the “tree” image “http://sipi.usc.edu/database/database.php?volume=misc&image=6#top”, the
image is first enhanced by the modified alpha-rooting and, then, the enhanced image is passed
through the power law (or gamma) transformation. Enhancement results are shown in Figure 3.
Figure 3:(a) The surface plot of the CEME vs `and k for i = 0.33 for the image “Tree[36]
” -
*
http://sipi.usc.edu/database/database.php?volume=misc&image=6#top; (b) The original Image “tree*
”; (c)
the histogram of the original image; (d) the enhanced image by the method of 2-DQDFT Modified Alpha-
Rooting; (e) the histogram of the image is (d);
Figure3:(continued) (f) The gamma transformation of the image in (d) with gamma=1.15;(g) the histogram
of the image in (f).
The “tree” image enhanced by combined enhancement method of modified alpha-rooting
followed by the gamma rooting (Figure 4b) is compared with enhancement of “tree” image with
the retinex algorithm (Figure 4c) and the gamma transformation method (Figure 4d).

Figure 4: (a) Original image “tree
*
http://sipi.usc.edu/database/database.php?volume=misc&image=6#top
The comparison of images in Figure
combined effect of modified alpha
much better enhanced image when compared to original (
retinex method (Figure 4c), or
(Figure 4d).
Figure 4: (continued) (b) The gamma transformation of the image enhanced by the modified alpha rooting
method; (c) The color image enhancement by
transformation alone of the original image; (e)
Figure5:(a) Surface plot of CEME vs
*https://dragon.larc.nasa.gov/retinex/pao/news/lg
(a)
Figure 4: (a) Original image “tree[36]
”-
http://sipi.usc.edu/database/database.php?volume=misc&image=6#top ;
Figure 4(a)-(e) shows that the enhancement of the image by the
combined effect of modified alpha-rooting followed by the gamma rooting (Figure
image when compared to original (Figure 4a), or the enhancement by
or the enhancement by the gamma transformation method alone
amma transformation of the image enhanced by the modified alpha rooting
mage enhancement by the retinex algorithm (McCann-99); (d) The gam
transformation alone of the original image; (e) The histogram equalized image.
Figure5:(a) Surface plot of CEME vs i, λ for `=0.98 for the image “Girl in the Blue Truck –
https://dragon.larc.nasa.gov/retinex/pao/news/lg-image16.jpg
9
the enhancement of the image by the
Figure 4b) gave a
enhancement by
the gamma transformation method alone
amma transformation of the image enhanced by the modified alpha rooting
99); (d) The gamma
istogram equalized image.
image16[37]
” -

In Figure 5, the original image “Girl in th
https://dragon.larc.nasa.gov/retinex/pao/news/lg
alpha-rooting and then followed by the
The value of i, and λ are obtained from the surface plot of CEME versus
`=0.98. The resulting enhanced images and histograms are shown in Figure 5 (b to g).
Figure 5:(continued) Surface plot of CEME vs
image16[37]
” - *https://dragon.larc.nasa.gov/retinex/pao/news/lg
in the Blue Truck – image16*
”; (c) The histogram of the original image; (d) The enhanced image by the
method of the 2-DQDFT Modified Alpha
transformed image of the image in (d) for
The enhanced image by the combined effect of
gamma transformation can be compared with the enhanced image by
enhancement by the gamma transformation alone.
image “Girl in the Blue Truck –
transformation was better, than the original image, or the enhancement by
or by the gamma transformation alone.
(b)
In Figure 5, the original image “Girl in the Blue Truck – image16
https://dragon.larc.nasa.gov/retinex/pao/news/lg-image16.jpg was enhanced by the modified
rooting and then followed by the gamma transformation.
are obtained from the surface plot of CEME versus i, and λ for the given
=0.98. The resulting enhanced images and histograms are shown in Figure 5 (b to g).
) Surface plot of CEME vs i, λ for `=0.98 for the image “Girl in the Blue Truck
https://dragon.larc.nasa.gov/retinex/pao/news/lg-image16.jpg (b) The original Image “Girl
DQDFT Modified Alpha-Rooting; (e) The histogram of the image in (d);
transformed image of the image in (d) for = 1 and gamma=1.258; (g) histogram of the image in (f).
The enhanced image by the combined effect of the modified alpha-rooting followed by the
compared with the enhanced image by the retinex algorithm and
amma transformation alone. The results show that the enhancement of the
image16” by the modified alpha-rooting followed by
than the original image, or the enhancement by the retinex
or by the gamma transformation alone.
(a)
(c)
10
image16 [37]
” -
image16.jpg was enhanced by the modified
, and λ for the given
=0.98. The resulting enhanced images and histograms are shown in Figure 5 (b to g).
=0.98 for the image “Girl in the Blue Truck –
(b) The original Image “Girl
Rooting; (e) The histogram of the image in (d);(f) Gamma
and gamma=1.258; (g) histogram of the image in (f).
rooting followed by the
algorithm and
esults show that the enhancement of the
rooting followed by the gamma
retinex algorithm,

(d)
Figure 6: (a) The original Image “Girl in the Blue Truck
*
*https://dragon.larc.nasa.gov/retinex/pao/news/lg
image enhanced by the modified alpha
algorithm (McCann-99);(d) The gamma transformation alone of the original image; (e)
equalized image of the original image.
In
“http://sipi.usc.edu/database/database.php?volume=misc&image=6#top”, the R
7a) was first converted to the
techniques of the modified alpha
the original XYZ image (Figure
obtained by the processing of the original RGB image (
After processing, the image in Figure
CEME measure of the image in Figure
in Figure 3f, where the processing is done directly on
Figure 7: (a) The original RGB Image “Tree
*
converted to the XYZ color model;
alpha rooting method done on the image in (b); (d) The image in (c) is conver
2.1. Using the 2-D DFT
The color image enhancement by
corresponding to the EME vs `
(e)
Figure 6: (a) The original Image “Girl in the Blue Truck – image16[37]
”-
https://dragon.larc.nasa.gov/retinex/pao/news/lg-image16.jpg ;(b) The gamma transformation of the
image enhanced by the modified alpha-rooting method; (c) The color image enhancement by the retinex
(d) The gamma transformation alone of the original image; (e) Theh
equalized image of the original image.
Figure7(a-e),
“http://sipi.usc.edu/database/database.php?volume=misc&image=6#top”, the RGB image (
the XYZ color model (Figure 7b). Then, the image enhancement
modified alpha-rooting followed by the gamma transformation we
Figure 7c). The CEME measure is found to be greater, than the value
obtained by the processing of the original RGB image (Figure 3f).
Figure 7c was converted back to the RGB image (Figure
Figure 7d is low as compared to the CEME measure of the image
where the processing is done directly on the RGB image.
Figure 7: (a) The original RGB Image “Tree[36]
”-
u/database/database.php?volume=misc&image=6#top ; (b) The RGB image in (a) is
converted to the XYZ color model; (c) Thegamma transformation of the image enhanced by the modified
alpha rooting method done on the image in (b); (d) The image in (c) is converted back to the
color image enhancement by modified alpha-rooting by the 2-D DFT of the image “Tree
http://sipi.usc.edu/database/database.php?volume=misc&image=6#topwas done. The surfaces
and λ for i = 0.33 for all three channels were plotted
11
(b) The gamma transformation of the
by the retinex
Thehistogram
“Tree”image
GB image (Figure
image enhancement
were appliedon
than the value
Figure 7d). The
the CEME measure of the image
; (b) The RGB image in (a) is
amma transformation of the image enhanced by the modified
the RGB image.
image “Tree*
”- *
The surfaces
re plottedseparately

12
(Figures 8(a to c)), to find the values of ` and λ which give a maximum EME measure. For
channel 1 (red), the maximum EME value is 35.5148 corresponding to `=0.74 and λ=0.14. For
channel 2 (green), the maximum EME value is 33.7420 corresponding to `=0.88 and λ=0.3. For
channel 3 (blue), the maximum EME value is 34.8439 corresponding to `=0.78 and λ=0.16.The
EME values of channels 1,2, and 3 of the original image equal 12.1228, 21.7077, and 12.1090,
respectively.
Figure8f shows the enhanced image composed of the channels that give the maximum EME
measures, when the channels were enhanced separately by the modified alpha-rooting 2-D DFT.
Then, Figure8h shows the enhanced image by the combined effects of the modified alpha-rooting
2-D DFT followed by the gamma transformation with γ = 3.51.
Figure 8 (a) The surface of the EME versus ` and λ for i = 0.33 for Channel 1 of the original image
“Tree[36]
”- *
http://sipi.usc.edu/database/database.php?volume=misc&image=6#top ;(b) The surface of the
EME versus ` and λ for i = 0.33 for Channel 2 for “tree” image;(c) The surface of EME vs ` and λ for i =
0.33 for Channel 3 for the “tree” image.;

Figure 8 (continued) (c) The surface of EME vs
(d) The original “Tree” image; (e) The channel b
channel enhancement by the modified alpha
corresponding to the maximum EME values; (g) The channel by channel histogram of the image in (f)
before scaling;(h) The channel by channel enhancement by
composed of enhanced channels corresponding to
transformation; (i) The channel by channel histogram of imag
In Figure 9(a)-(d), the modified alpha rooting followed by
accomplished (see Figure9c) in the
After the processing the image, it
λ values for the channels X, Y, and Z corresponding to the maximums of the EME measure are
respectively [0.78, 0.84, 0.78] and [0.26, 0.42, 0.16]. The EME measure of the original X, Y, and
Z channels are [13.0022, 18.1088, 12.1494] and the maximums
after enhancement are [34.8012, 35.0018, 34.6950], and when the processed in the XYZ image is
converted back to the RGB image, the corresponding EME measures of the channels R, G, and B
are [20.8039, 36.0317, 46.7360].
(c) The surface of EME vs ` and λ for i = 0.33 for Channel 3 for the “tree” image.;
(d) The original “Tree” image; (e) The channel by channel histogram of image in (d);(f) The channel
channel enhancement by the modified alpha-rooting, when the image is composed of enhanced channels
(h) The channel by channel enhancement by the modified alpha-rooting, when the
composed of enhanced channels corresponding to the maximum EME values – followed by
transformation; (i) The channel by channel histogram of image in (h) before scaling
modified alpha rooting followed by the gamma transformation
the XYZ image model (Figure9b) of the RGB image in
, itwas converted back to the RGB image (in Figure9
values for the channels X, Y, and Z corresponding to the maximums of the EME measure are
Z channels are [13.0022, 18.1088, 12.1494] and the maximums of the EME measure obtained
are [20.8039, 36.0317, 46.7360].
(a) (b)
13
= 0.33 for Channel 3 for the “tree” image.;
y channel histogram of image in (d);(f) The channel-by-
rooting, when the image is composed of enhanced channels
, when the image is
followed by the gamma
before scaling.
gamma transformation was
b) of the RGB image in Figure9a.
9d).The ` and
values for the channels X, Y, and Z corresponding to the maximums of the EME measure are
of the EME measure obtained

(c)
Figure 9: (a) The original RGB image “tree
*
converted to the XYZ color model;
channel by the method of modified alpha rooting method done; (d) The image in (c) is converted back to
4. CONCLUSIONS
The color image enhancement technique of the modified alpha
quaternion discrete Fourier transform shows better enhancement techniques as compared to
processing the image channel-
individual enhanced images. One of the main prominent features in channel
enhancement techniques is that the color obtained after the processing has marked
from that of the original image and also of the color of an image when processing by the pro
quaternion approach. The integrity of the color is lost in channel
processing in the quaternion approach preserves the uniqueness of the color. In future studies, the
human sensitiveness of color perception and the ps
to understand the advantage of the quaternion approach of color image enhancement techniques
over the channel-by-channel enhancement
REFERENCES
[1] R.C. Gonzalez and R.E. Woods, Digital Image Processing,
[2] A.M. Grigoryan, S.S. Again, A.M. Gonzales, “Fast Fourier transform
color image enhancement,” [9497
Security, and Applications, 20
[3] A. M. Grigoryan, S.S. Agaian, “2
Algorithm,” Proceedings, Electronic Imaging 2016, IS&T, February 14
[4] A.M. Grigoryan, S.S. Again, “Tensor representation of color images and fast 2
Fourier transform,” [9399-16], Proceedings of SPIE vol. 9399, 2015 Electronic Imaging: Image
Processing: Algorithms and Systems XIII, February 10
[5] A.M. Grigoryan and M.M. Grigoryan, Brief Notes in Advanced DSP: Fourier Analysis With
MATLAB, CRC Press Taylor and Francis Group, 2009.
[6] A.M. Grigoryan and S.S. Agaian, "Retooling of color imaging in the quaternion algebra," Applied
Mathematics and Sciences: An International Journal, vol. 1, no. 3, pp. 23
[7] A.M. Grigoryan and S.S. Agaian, “Transform
performance measure,” Advances in Imaging and Electron Physics, Academic Pre
165–242, May 2004.
[8] A.M. Grigoryan and S.S. Agaian, “Optimal color image restoration: Wiener filter and quaternion
Fourier transform,” [9411-24], Proceedings of SPIE vol. 9411, 2015 Electronic Imaging: Mobile
Devices and Multimedia: Enabling Technologies, Algorithms, and Applications 2015, February 10
11, San Francisco, California, 2015.
(d)
Figure 9: (a) The original RGB image “tree[62]
”-
http://sipi.usc.edu/database/database.php?volume=misc&image=6#top ; (b) The RGB image in (a) is
onverted to the XYZ color model;(c) The gamma transformation of the image in (b) enhanced channel by
the RGB image.
ncement technique of the modified alpha-rooting by the
-by-channel and then composing the color image from these
anced images. One of the main prominent features in channel
enhancement techniques is that the color obtained after the processing has marked
from that of the original image and also of the color of an image when processing by the pro
quaternion approach. The integrity of the color is lost in channel-by-channel processing, while the
human sensitiveness of color perception and the psychophysics of the color will be studied better
channel enhancement.
R.C. Gonzalez and R.E. Woods, Digital Image Processing, 2nd Edition, Prentice Hall, 2002.
A.M. Grigoryan, S.S. Again, A.M. Gonzales, “Fast Fourier transform-based retinex and alpha
color image enhancement,” [9497-29], Proc. SPIE Conf., Mobile Multimedia/Image Processing,
20 - 24 April 2015, Baltimore, Maryland United States.
A. M. Grigoryan, S.S. Agaian, “2-D Left-Side Quaternion Discrete Fourier Transform: Fast
Algorithm,” Proceedings, Electronic Imaging 2016, IS&T, February 14-18, San Francisco, 2016.
yan, S.S. Again, “Tensor representation of color images and fast 2-D quaternion discrete
16], Proceedings of SPIE vol. 9399, 2015 Electronic Imaging: Image
Processing: Algorithms and Systems XIII, February 10-11, San Francisco, California, 2015.
A.M. Grigoryan and M.M. Grigoryan, Brief Notes in Advanced DSP: Fourier Analysis With
MATLAB, CRC Press Taylor and Francis Group, 2009.
A.M. Grigoryan and S.S. Agaian, "Retooling of color imaging in the quaternion algebra," Applied
Mathematics and Sciences: An International Journal, vol. 1, no. 3, pp. 23-39, December 2014
A.M. Grigoryan and S.S. Agaian, “Transform-based image enhancement algorithms with
performance measure,” Advances in Imaging and Electron Physics, Academic Press, vol. 130, pp.
A.M. Grigoryan and S.S. Agaian, “Optimal color image restoration: Wiener filter and quaternion
24], Proceedings of SPIE vol. 9411, 2015 Electronic Imaging: Mobile
abling Technologies, Algorithms, and Applications 2015, February 10
11, San Francisco, California, 2015.
14
(b) The RGB image in (a) is
amma transformation of the image in (b) enhanced channel by
rooting by the two-side 2D
channel and then composing the color image from these
anced images. One of the main prominent features in channel-by-channel
enhancement techniques is that the color obtained after the processing has marked difference
from that of the original image and also of the color of an image when processing by the proposed
channel processing, while the
ychophysics of the color will be studied better
2nd Edition, Prentice Hall, 2002.
based retinex and alpha-rooting
29], Proc. SPIE Conf., Mobile Multimedia/Image Processing,
Side Quaternion Discrete Fourier Transform: Fast
18, San Francisco, 2016.
D quaternion discrete
16], Proceedings of SPIE vol. 9399, 2015 Electronic Imaging: Image
lifornia, 2015.
A.M. Grigoryan and M.M. Grigoryan, Brief Notes in Advanced DSP: Fourier Analysis With
A.M. Grigoryan and S.S. Agaian, "Retooling of color imaging in the quaternion algebra," Applied
39, December 2014
based image enhancement algorithms with
ss, vol. 130, pp.
A.M. Grigoryan and S.S. Agaian, “Optimal color image restoration: Wiener filter and quaternion
24], Proceedings of SPIE vol. 9411, 2015 Electronic Imaging: Mobile
abling Technologies, Algorithms, and Applications 2015, February 10-

15
[9] A.M. Grigoryan, J. Jenkinson and S.S. Agaian, “Quaternion Fourier transform based alpha-rooting
method for color image measurement and enhancement,” SIGPRO-D-14-01083R1, Signal Processing,
vol. 109, pp. 269-289, April 2015, (doi: 10.1016/j.sigpro.2014.11.019).
[10] S.S. Agaian, K. Panetta, and A.M. Grigoryan, “A new measure of image enhancement, in Proc. of the
IASTED Int. Conf. Signal Processing Communication, Marbella, Spain, Sep. 1922, 2000.
[11] T.A. Ell, N. Bihan and S.J. Sangwine, “Quaternion Fourier transforms for signal and image
processing,” Wiley NJ and ISTE UK, 2014
[12] W.R. Hamilton, Elements of Quaternions, Logmans, Green and Co., London, 1866.
[13] T.A. Ell, “Quaternion-Fourier transforms for analysis of 2-dimensional linear time-invariant partial
differential systems,” In Proceedings of the 3nd IEEE Conference on Decision and Control, vol. 1-4,
pp. 1830–1841, San Antonio, Texas, USA, December 1993.
[14] S.J. Sangwine, “Fourier transforms of colour images using quaternion, or hypercomplex, numbers,”
Electronics Letters, vol. 32, no. 21, pp. 1979–1980, October 1996.
[15] S.J. Sangwine, “The discrete quaternion-Fourier transform,” IPA97, Conference Publication no. 443,
pp. 790–793, July 1997.
[16] S.J. Sangwine, T.A. Ell, J.M. Blackledge, and M.J. Turner, “The discrete Fourier transform of a color
image,” in Proc. Image Processing II Mathematical Methods, Algorithms and Applications, pp. 430–
441, 2000.
[17] S.J. Sangwine and T.A. Ell, “Hypercomplex Fourier transforms of color images,” in Proc. IEEE Intl.
Conf. Image Process., vol. 1, pp. 137-140, 2001
[18] S.S. Agaian, K. Panetta, and A.M. Grigoryan, (2001) “Transform-based image enhancement
algorithms,” IEEE Trans. on Image Processing, vol. 10, pp. 367-382.
[19] A.M. Grigoryan and S.S. Agaian, (2014) “Alpha-rooting method of color image enhancement by
discrete quaternion Fourier transform,” [9019-3], in Proc. SPIE 9019, Image Processing: Algorithms
and Systems XII, 901904; 12 p., doi: 10.1117/12.2040596
[20] J.H. McClellan, (1980) “Artifacts in alpha-rooting of images,” Proc. IEEE Int. Conf. Acoustics,
Speech, and Signal Processing, pp. 449-452.
[21] F.T. Arslan and A.M. Grigoryan, (2006) “Fast splitting alpha-rooting method of image enhancement:
Tensor representation,” IEEE Trans. on Image Processing, vol. 15, no. 11, pp. 3375-3384.
[22] A.M. Grigoryan and F.T. Arslan, "Image enhancement by the tensor transform," Proceedings, IEEE
International Symposium on Biomedical Imaging: from Macro to Nano, ISBI 2004, pp. 816-819,
April 2004, Arlington, VA.
[23] A.M. Grigoryan and S.S. Agaian, “Cell Phone Cameral Color Medical Imaging via Fast Fourier
Transform,” in Electronic Imaging Applications in Mobile Healthcare, J. Tang, S.S. Agaian, and J.
Tan, Eds., SPIE Press, Bellingham, Washington, pp. 33-76 (2016).
[24] A.M. Grigoryan and S. Dursun, "Multiresolution of the Fourier transform," in the Proceedings of the
IEEE International Conference on Acoustics, Speech, and Signal Processing, 2005 (ICASSP '05), vol.
4, pp. 577–580, March 18-23, 2005, Philadelphia, PA
[25] A.M. Grigoryan and S.S. Agaian, “Tensor form of image representation: Enhancement by image-
signals," [5014-26] (Image Processing: Algorithms and Systems II, Electronic Imaging, Science and
Technology 2003, IS&T/SPIE 15 Annual Symposium, Santa Clara, CA, January 20-24, 2003)
[26] A.M. Grigoryan, “2-D and 1-D multi-paired transforms: Frequency-time type wavelets,” IEEE Trans.
on Signal Processing, vol. 49, no. 2, 344–353 (2001)
[27] S.S. Agaian, H.G. Sarukhanyan, K.O. Egiazarian, J. Astola, Hadamard transforms, SPIE Press, 2011.
[28] A.M. Grigoryan, K. Naghdali, “On a method of paired representation: Enhancement and
decomposition by series direction images,” Journal of Mathematical Imaging and Vision, vol. 34, no.
2, 185–199 (2009)
[29] F.T. Arslan and A.M. Grigoryan, "Enhancement of medical images by the paired transform," in
Proceedings of the 14th IEEE International Conference on Image Processing, vol. 1, pp. 537-540, San
Antonio, Texas, September 16-19, 2007.
[30] F.T. Arslan, J.M. Moreno, and A.M. Grigoryan, "Paired directional transform-based methods of
image enhancement," in Proceedings of SPIE – vol. 6246, Visual Information Processing XV, April
18-19, 2006.
[31] A.M. Grigoryan and S.S. Agaian, “Color Enhancement and Correction for Camera Cell Phone
Medical Images Using Quaternion Tools,” in Electronic Imaging Applications in Mobile Healthcare,
J. Tang, S.S. Agaian, and J. Tan, Eds., SPIE Press, Bellingham, Washington, pp. 77-117 (2016).

16
[32] A.M. Grigoryan, S.S. Agaian, “Alpha-Rooting Method of Gray-scale Image Enhancement in The
Quaternion Frequency Domain,” Proceedings of IS&T International Symposium, Electronic Imaging:
Algorithms and Systems XV, 29 Jan.-2 Feb., Burlingame, CA, 2017.
[33] A.M. Grigoryan, A. John, S.S. Agaian, “Color image enhancement of medical images using alpha-
rooting and zonal alpha-rooting methods on 2-D QDFT,” Proceedings of SPIE Medical Imaging
Symposium, Image Perception, Observer Performance, and Technology Assessment, 12 - 13 Feb.,
Orlando, FL, 2017.
[34] http://www.nasa.gov/mission_pages/mars/images/index.html
[35] http://sipi.usc.edu/database/database.php?volume=misc&image=6#top
[36] https://dragon.larc.nasa.gov/retinex/pao/news/lg-image16.jpg
AUTHORS
Artyom M. Grigoryan received the PhD degree in Mathematics and Physics from
Yerevan State University (1990). He is an associate Professor of the Department of
Electrical Engineering in the College of Engineering, University of Texas at San Antonio.
The author of four books, 9 book-chapters, 3 patents and more than 120 papers and
specializing in the design of robust filters, fast transforms, tensor and paired transforms,
discrete tomography, quaternion imaging, image encryption, processing biomedical
images.
Aparna John received her B. Tech. degree in Applied Electronics and Instrumentation
from University of Calicut, India and M. Tech. degree in Electronics and Communication
with specialization in Optoelectronics and Optical Communication from University of
Kerala, India. Now, she is a doctoral student in Electrical Engineering at University of
Texas at San Antonio. She is pursuing her research under the supervision of Dr. Artyom
M. Grigoryan. Her research interests include image processing, color image enhancements, fast algorithms,
quaternion algebra and quaternion transforms including quaternion Fourier transforms. She is a student
member of IEEE and also a member of Eta Kappa Nu Honor Society, University of Texas San Antonio
Sos S. Agaian is the Distinguished Professor at the City University of New York/CSI. Dr.
Agaian received the Ph.D. degree in math and physics from the Steklov Institute of
Mathematics, Russian Academy of Sciences, and the Doctor of Engineering Sciences
degree from the Institute of the Control System, Russian Academy of Sciences. He has
authored more than 500 scientific papers, 7 books, and holds 14 patents. His research
interests are Multimedia Processing, Imaging Systems, Information Security, Artificial Intelligent,
Computer Vision, 3D Imaging Sensors, Fusion, Biomedical and Health Informatics.

MODIFIED ALPHA-ROOTING COLOR IMAGE ENHANCEMENT METHOD ON THE TWO-SIDE 2-DQUATERNION DISCRETE FOURIER TRANSFORM AND THE 2-DDISCRETE FOURIER TRANSFORM

More Related Content

What's hot (17)

Similar to MODIFIED ALPHA-ROOTING COLOR IMAGE ENHANCEMENT METHOD ON THE TWO-SIDE 2-DQUATERNION DISCRETE FOURIER TRANSFORM AND THE 2-DDISCRETE FOURIER TRANSFORM (20)

More from mathsjournal (20)

Recently uploaded (20)

MODIFIED ALPHA-ROOTING COLOR IMAGE ENHANCEMENT METHOD ON THE TWO-SIDE 2-DQUATERNION DISCRETE FOURIER TRANSFORM AND THE 2-DDISCRETE FOURIER TRANSFORM