Received 23 April 2023, accepted 21 May 2023, date of publication 24 May 2023, date of current version 7 June 2023.

Digital Object Identifier 10.1109/ACCESS.2023.3279734

Color to Grayscale Image Conversion Based on

Singular Value Decomposition
ZAID NIDHAL KHUDHAIR 1,4 , AHMED NIDHAL KHDIAR 2 , NIDHAL K. EL ABBADI 3,

FARHAN MOHAMED 4 , (Senior Member, IEEE), TANZILA SABA 5 ,

FATEN S. ALAMRI 6 , AND AMJAD REHMAN 5 , (Senior Member, IEEE)
1 Faculty of Engineering, School of Computing, University of Technology Malaysia, Johor Bahru 81310, Malaysia
2 Department of Electrical Engineering, Faculty of Engineering, University of Kufa, Najaf 54001, Iraq
3 Computer Techniques Engineering Department, Al-Mustaqbal University, Babylon 51001, Iraq
4 UTM-IRDA MaGICX, Institute of Human Centered Engineering, Universiti Teknologi Malaysia, Johor Bahru 81310, Malaysia
5 Artificial Intelligence and Data Analytics Laboratory, College of Computer and Information Sciences (CCIS), Prince Sultan University, Riyadh 11586,

Saudi Arabia
6 Department of Mathematical Sciences, College of Science, Princess Nourah bint Abdulrahman University, Riyadh 11671, Saudi Arabia

Corresponding author: Tanzila Saba ([email protected]) and Faten S. Alamri ([email protected])

This research was funded by Princess Nourah bint Abdulrahman University and Researchers Supporting Project number
(PNURSP2023R346), Princess Nourah bint Abdulrahman University, Riyadh, Saudi Arabia.

ABSTRACT Color information is useless for distinguishing significant edges and features in numerous
applications. In image processing, a gray image discards much-unrequired data in a color image. The
primary drawback of colour-to-grey conversion is eliminating the visually significant image pixels. A current
proposal is a novel approach for transforming an RGB image into a grayscale image based on singular value
decomposition (SVD). A specific factor magnifies one of the color channels (Red, Green, and Blue). A vector
of three values (Red, Green, Blue) of each pixel in an image is decomposed using SVD into three matrices.
The norm of the diagonal matrix was determined and then divided by a specific factor to obtain the grey
value of the corresponding pixel. The contribution of the proposed method gives the user high flexibility to
produce many versions of gray images with varying contrasts, which is very helpful in many applications.
Furthermore, SVD allows for image reconstruction by combining the weighting of each channel with the
singular value matrix. This results in a grayscale image that more accurately captures the actual intensity
values of the image and preserves more color information than traditional grayscale conversion methods,
resulting in loss of color information. The proposed method was compared with a similar method (converting
the color image into grayscale) and was found to be the most efficient.

INDEX TERMS Decolorization, grey image, image conversion, SVD, technological development.

I. INTRODUCTION ranged from zero (black) to 255 (white), as shown in Fig. 1.

Currently, most of the captured images are color images. Generally, the conversion process produces an image with
However, it is frequently necessary to convert a color image to lower contrast than the original color image. The goal of each
grayscale images for printing, aesthetic intents, object detec- conversion algorithm is to preserve the local luminance con-
tion, and publishing as less expensive, helping colorblind sistency, global consistency, image contrast information, and
people preserve visual cues [1]. Color-to-grayscale conver- hue order as much as possible [3]. Different processing meth-
sion is the process of reducing the image dimensions by ods are required for each pixel in various image-processing
transforming the RGB tristimulus values (Red, Green, Blue) applications. Processing RGB pixels is not feasible due to
∈ R3 to the intensity value (I) R [2]. The lightness values high storage requirements and computation costs. The best
solution to these issues is to convert an RGB image into a
The associate editor coordinating the review of this manuscript and grayscale image [4]. Grayscale images are mainly used for
approving it for publication was Mingbo Zhao . shape characteristics, edge detection, circular objects, and
This work is licensed under a Creative Commons Attribution-NonCommercial-NoDerivatives 4.0 License.
VOLUME 11, 2023 For more information, see https://creativecommons.org/licenses/by-nc-nd/4.0/ 54629
 Z. N. Khudhair et al.: Color to Grayscale Image Conversion Based on SVD

works is introduced. The methodology is presented in Section

six. Section seven focuses on the results and validates the
proposed algorithm. Finally, we conclude this reserach in
section eight.

II. PROBLEM DESCRIPTION

Although many conversion techniques have been developed
FIGURE 1. Sample of the grayscale image.
and reported in state of the art, they still pose several critical
problems, such as [8] and [9]:
• Color distortion: Many grayscale conversion methods
corners. Illumination and chrominance for matching and cor-
can result in color distortion. This is because these meth-
relation had negative effects. Grey images are highly useful
ods cannot accurately capture an image’s true intensity
for further segmentation. The contrast and smoothness can
values.
be increased in grey images to obtain more image details.
• Loss of color information: Many grayscale conversion
A grey image requires less memory space and is processed
methods result in the loss of color information, which
faster, which is common in image processing.
can be significant in cases where the color information
Suggesting a good algorithm that can transform an RGB
is important for interpreting the image.
image into a gray image while preserving salient features is a
• Reduced image quality: Conversion of color images into
significant challenge and a complicated process. During the
grayscale can reduce image quality, particularly in cases
conversion process, numerous crucial image features, such as
where the image contains subtle color variations that are
sharpness, shadows, contrast, and color structure, may need
important for interpretation.
to be recovered. In addition, current methods consume large
• Inadequate representation: In some cases, grayscale
amounts of computational time and memory [5].
images may not adequately represent the original color
Algorithms for converting images into grayscale can be
image because the information in the color channels may
classified into two major groups. The first is based on color
be lost during the conversion process.
transformation. Usually, the color image (RGB) is transferred
• High computational complexity: Some grayscale con-
into other color spaces, separating the luminance from the
version methods, such as those based on the color
chrominance channels. In the second group, the lightness
appearance model, can be computationally expensive,
channel can be extracted from the color space (L∗ a∗ b∗ ) to
rendering them unsuitable for real-time application.
build a grayscale image. The main problem with current color
transform methods is that mapping the same gray value for These problems have motivated researchers to develop new
various color values with the same luminance leads to the loss methods to convert color images into grayscale images that
of edge and object boundaries [6]. are more accurate, efficient, and robust. The SVD method
Methods of converting color images to grayscale can be aims to preserve important color information in an image,
classified into three types: global method which has the power reduce color distortion, and enhance the image’s visual qual-
to produce a natural-looking gray image, in general, it is ity.
time-consuming. The second type is the local method, which
has good local contrast preservation but may produce a local III. ADVANTAGES OF THE PROPOSED METHOD
artifact. In addition, a poor property of the local mapping The proposed method for converting color images to
method is the possibility of mapping the same color value grayscale is highly efficient and does not require specialized
into various gray values. The third type is hybrid mapping, hardware, making it well suited for practical applications. It is
which combines the two previous mappings [7]. designed to be easy to use and integrate into existing image
Using SVD to convert a color image to a grayscale enables processing pipelines.
the representation of the image as a collection of singular In addition, the proposed method allows users to obtain
values that capture the intensity information of the image. high-quality grayscale images without sacrificing the critical
This can be useful in image analysis and processing tasks details contained in color channels. This is achieved through
performed on grayscale images. In addition, SVD is a robust a novel weighting scheme that effectively balances the con-
technique for image processing, making it suitable for pro- tributions of different color channels to create a grayscale
cessing images that may be degraded or contaminated by image that accurately reflects the underlying color content
noise. of the original image. As a result, this method can be used
The remaining paper is structured as follows: Section two confidently in a wide range of applications in which color
described the main problems when converting color to a information is essential for accurate analysis and interpreta-
grayscale image. Whereas section three provides the main tion.
advantages of the current proposal. Section four provides a An important advantage of our grayscale conversion
brief overview of prior research on converting color images method is that it avoids color distortion, which can be a
to grayscale. In section five an idea about SVD and how it significant problem in many existing approaches.

54630 VOLUME 11, 2023

Z. N. Khudhair et al.: Color to Grayscale Image Conversion Based on SVD

result, the benefits and drawbacks of color-to-grayscale con-

version remain unknown, some of which are:
The authors in [8] presented a method that converts a
color image into other color spaces (L∗ a∗ b∗ ), where the
luminance and chrominance components are separated and
processed independently. The chrominance information is
decomposed using singular value decomposition (SVD). The
weight matrix was constructed by multiplying the eigenval-
ues and eigenvectors of the chrominance planes produced
from the SVD analysis. To obtain low-contrast information,
luminance information is combined with a weighted sum
of chrominance data. This method has several advantages
related to the preservation of color information, robustness,
FIGURE 2. Conversion of a color image by three different methods.
and flexibility, as well as potential disadvantages related to
complexity, subjectivity, and loss of fine details.
In [14], a color-to-grayscale transformation employed
One of the key advantages of the current method is its
three global linear weighting parameters derived directly
robustness to noise, which enables it to perform well even in
from the correlations between the three-color channels (red,
noisy or low-quality images. In contrast, other methods may
green, and blue) concerning a base image. The author con-
need help to produce high-quality grayscale results.
siders the correlation values (magnitude and sign) to set the
weighting parameters. This method has several advantages,
IV. BACKGROUND
including its efficiency, direct RGB channel processing, and
Several methods have been proposed for converting color
comparability to state-of-the-art methods. However, it also
images into grayscale images. One of these methods is to use
has limitations, including limited applicability to other image
one of the (red, green, or blue) channels of the RGB image
processing tasks, limited accuracy due to reliance on corre-
or one channel of another image color space, such as using
lations between channels and a contrast image, and limited
channel V of the HSV color space. These methods may cause
flexibility due to fixed linear weighting parameters.
a loss of discriminative information [2].
In [15], a method for converting a color image into a
A very simple method is based on the mean of the maxi-
grayscale image was proposed based on extracting eigenval-
mum and minimum values of three colors called (the lightness
ues and eigenvectors from SVD. In this study, converting
method) as in ‘‘equation (1)’’ [10]. The major weakness of
the input color image to the CIE L∗a∗b∗ space is beneficial
this method is that it does not use three-color components.
for separately processing the luminance and chrominance
min(R, G, B) + Max(R, G, B) information.
Grayscale image = (1)
2 Multiplying the eigenvalues and eigenvectors resulting
Another method obtained the grayscale value from the aver- from the application of SVD to CIELab∗ produces chromi-
age value of the three colors as in ‘‘equation (2)’’ [11], [12]: nance matrices. The weighted sum of chrominance and lumi-
nance information obtained from the grayscale image. This
R+G+B
Grayscale image = (2) method has several advantages related to the preservation of
3 color information, robustness, and flexibility, it also has some
This method assigns the same weight to each of the three- potential disadvantages related to complexity, subjectivity,
color components, which is a weakness because our eyes and loss of fine details.
react differently to each color. In [16], a CNN incorporating a local feature network
One of the best methods for converting color to gray is and coarse classifier was proposed to differentiate between
based on a weighted combination of RGB channels, called different exposure conditions of color images. The network
the luminosity method, as shown in ‘‘equation (3)’’ [13]: accomplishes this by learning local semantic features. The
Grayscale = 0.2989 × R + 0.587 × G + 0.114 × B (3) authors aimed to achieve optimal results in preserving the
color contrast and exposure settings by training the model
However, this method has several areas for improvement, to map color and grayscale images. This method has some
including a lack of color separation, insensitivity to satura- potential advantages in preserving color contrast and expo-
tion, poor handling of certain color combinations, and limited sure settings as well as some potential weaknesses related to
dynamic range. These weaknesses can result in a loss of dependence on training data, complexity, loss of fine details,
contrast, detail, and color separation in the resulting grayscale and difficulty in preserving color contrast.
image. In [17], the color image (RGB) is separated into three
The differences between the two methods are shown in channels (red, green, and blue) based on the information-
Fig. 2. Many conversion methods have recently been pro- theoretic method, and two weight maps are defined. Using
posed, but their performances still need to be assessed. As a information maximization, the first weight map extracts the

VOLUME 11, 2023 54631

 Z. N. Khudhair et al.: Color to Grayscale Image Conversion Based on SVD

that satisfies the following three conditions:

□ ∥v∥ ≥ 0 for all v ϵ V and
∥v∥ = 0 if and only if v = 0
□ ∥kv∥ = |k| ∥v∥ for any scalar k ϵ R
and vector v ϵ Rd
□ ∥v + w∥ ≤ ∥v∥ + ∥w∥ for any two
vectors v, w ϵ V
The norm of a vector can be thought of as the length or
FIGURE 3. Schematic diagram of SVD [22].
magnitude of the vector.

VI. BLIND IMAGE QUALITY ASSESSMENT

visually salient regions. By contrast, the second weight map Image quality assessment is a major challenge in digital
determines the local entropy per patch by filtering the local image processing. It can be classified as subjective (based
variation value for each image. This approach employs a on the perceptual assessment of a human viewer) or objec-
multiscale fusion technique to alleviate the local distortions tive (based on computational models and algorithms used to
caused by weight map discontinuities. Information-theoretic mimic the subjective assessment).
methods have several advantages related to the preservation No-reference image quality is the main problem in mea-
of visually salient regions, local contrast enhancement, and suring image quality, which is called blind image quality,
multi-scale fusion. However, the method may be computa- which has no prior knowledge about the distortion type.
tionally expensive and require additional optimization for Assessing the quality of an image without a reference has
certain image types or applications. long been a challenging research problem, as the image may
be subject to various types of distortion, which can cause
V. SINGULAR VALUE DECOMPOSITION (SVD)
highly varied image content. Additionally, the growing num-
In recent years, researchers in the field of image processing ber and significance of digital images in our daily lives have
have focused on using one of the mathematical algorithms resulted in significant interest within the research community
in the field of numerical linear algebra, which was applied in image quality assessment [23]. The Natural Image Quality
in a small number of fields at the emerging time but has Evaluator (NIQE) and Psychovisually-based image quality
rapidly expanded to many other fields. This algorithm uses evaluator (PIQE) are prominent techniques in this category.
singular-value decomposition (SVD). The NIQE assesses deviations from statistical regularities in
The SVD provides a capability for the researcher to use natural images without any exposure to distortion or training
a lot of various dimension matrices by a generalization of on human-rated distorted images. NIQE determines the per-
the eigenvalue decomposition idea [18]. SVD transforms ceptual quality of an image by comparing it to a default model
a matrix into three matrices U6VT (U and V are rotation derived from natural-scene images. A higher-quality image is
matrices and 6 is a scaling matrix) that carry some of the indicated by a lower score [24].
useful and unique properties of the original matrix. The SVD The Psychovisually-based image quality evaluator (PIQE)
method in image processing regards algebraic features [19]. demonstrated an inverse correlation with the perceptual qual-
SVD decomposes any (m, n) matrix into three matrices as ity of an image. A high score indicates low perceptual quality,
shown in Fig. 3, where m is the matrix rows, and n represents whereas a low score indicates high quality [25]. In the current
the matrix columns, where the rank r satisfies r ≤ n ≤ m, work, we used NIQE and PIQE to compare the quality of
it can be decomposed into three matrices as follows: color image conversion methods; in general, the NIQE and
A = USV T (4) PIQE values did not precisely reflect the quality of images,
but they could be approximate values because we did not
Recently SVD has gained popularity and become an attractive enhance the images before conversion (images may suffer
algebraic transform in image processing, owing to its stability from noise, blurring, etc.).
and conceptual advantages. SVD is a reliable and robust
orthogonal matrix factorization technique with prominent
properties in imaging [20]. Recall that the Frobenius norm VII. PROPOSED METHODOLOGY
of an n × m matrix A ∈ Rn×m is defined. The color-to-grayscale conversion method proposed in this
However, this norm does not directly correspond to A as a study is based on SVD, which is a mathematical technique
representation of a linear map. Specifically, in many contexts, that can be used to decompose a color image into its con-
measuring the difference between two matrices is essential stituent color channels. By considering the relative impor-
based on their effect on vectors. A norm on a vector space V tance of each channel, SVD can produce a grayscale image
is a function [21]: that more accurately captures the true intensity values of the
image and reduces color distortion. The main steps of this
∥.∥ V V → R proposal are summarized in Algorithm 1.

54632 VOLUME 11, 2023

Z. N. Khudhair et al.: Color to Grayscale Image Conversion Based on SVD

The goal of this algorithm is to convert a color image Algorithm 1: Converting Color Image Into a Grayscale Image
into a grayscale image. Grayscale images have only a single Input: color image
channel of information, where each pixel is represented by Output: grayscale image (Gray Image)
a single gray value instead of three-color values (red, green, Step 1: Input color image (X)
and blue), as in a color image. Step 2: Determine image size (number of rows (NR), number
The algorithm takes a color image (X) as input and sepa- of columns (NC)).
rates it into three color channels (red, green, and blue) using Step 3: Input parameter (k)
the RGB color model. The image size (the number of rows Step 4: Loop i = 1: NR
and columns) was determined. Step 5: Loop j = 1: NC
A loop is then used to iterate through each pixel in the Step 6: Read pixel (X (i, j))
image, starting at the top-left corner (i = 1, j = 1) and Step 7: Separate (X(i, j)) into three channels (Red (xr), Green
progressing row-by-row. (xg), and Blue (xb))
For each pixel, the algorithm creates a vector (C) by read- Step 8: Create a vector for each pixel
ing one byte from each of the three color channels (red, green, C (i, j) = [xr (i, j), xg (i, j), xb (i, j)]
and blue) at the current pixel location (i, j). Optionally, the Step 9: Add weight to one of the three parameters in the
algorithm can apply weight to one of the color channels to vector
create three different vectors (C1, C2, and C3) that can be C (i, j) = [3 × xr (i, j), xg (i, j), xb (i, j)]
used to create three different grayscale images. // Optional: you can create three vectors with different loca-
Singular Value Decomposition (SVD) was then applied tions of weight to get three different gray images
to each vector (C, C1, C2, and C3) separately. SVD is a C1 (i, j) = [3 × xr (i, j), xg (i, j), xb (i, j)]
mathematical process that can decompose a matrix into three C2 (i, j) = [xr (i, j), 3 × xg (i, j), xb (i, j)]
matrices (U, S, and V). These matrices represent the orig- C3 (i, j) = [xr (i, j), xg (i, j), 3 × xb (i, j)]
inal matrix’s rotation, scaling, and reflection. In this step, Step 10: Find the SVD
we focused on the diagonal matrix (S) and neglected the [U S V] = SVD (C (i, j))
other two matrices. The norm value of the diagonal matrix Step 11: Find the gray value
is determined using ‘‘equation (5).’’ G = norm (S) / k
Step 12: Save the result in the gray image
q
S⃗ = s12 +s22 + · · · + sn2 (5)
Gray Image (i, j) = G
where (s) is the value of elements in the diagonal of the Step 13: End loop // (step 5)
(S) matrix and (n) is the number of elements in a diagonal Step 14: End loop // (step 4)
matrix (rank). Step 15: Display the image.
The norm value (which may be larger than the image range
values (255)) is normalized by dividing it by a specific value
(k) to obtain the gray value corresponding to the color pixel.
Parameter (k) controls the degree of contrast in the resulting
grayscale image in addition to normalization.
The algorithm then calculates the gray value (G) for each
pixel using the formula

G = S⃗ /k k = 2, 3 . . . .N (6)

where norm(S) is the Frobenius norm of matrix S. The Frobe-

nius norm is a measure of the magnitude of the matrix S.
The gray value (G) was then saved in a grayscale image at
the corresponding pixel location (i, j). This process continues
until every pixel in the image has been processed.
Finally, the resulting grayscale image was displayed, show-
ing the original image in a single-channel gray format.

VIII. RESULTS ANALYSIS AND DISCUSSION

FIGURE 4. Comparing the converting color images into grayscale using
Initially, the proposed model was tested to convert color the Matlab with the proposed method. (multiply the red channel by three,
images into gray images. Accordingly, the results are shown and divide the norm by two).
in Fig. 4. We tested the effect of adding weight to one of
the three channels (multiplying by three, while the other
two channels did not change). In Fig. 5, the color image is by three, as shown in the second column. In contrast, the third
converted into a gray image by multiplying the red channel column shows the result of multiplying the green channel

VOLUME 11, 2023 54633

 Z. N. Khudhair et al.: Color to Grayscale Image Conversion Based on SVD

FIGURE 5. The difference between images when adding weight for each
channel (multiply by three), and simple images (dividing parameter equal
two).

by three, and finally, the blue channel is multiplied by three,

as shown in the fourth column.
From Fig. 5, we argue that adding weight to one of the
three channels may significantly affect the grayscale image
FIGURE 6. The result of color conversion is when multiplying each color
resulting from this method. However, some of the images channel by a specific weight (Dividing parameter equal to two.).
have a little effect depending on the image content. This will
help produce a gray image suitable for some applications.
The other test examines the effect of multiplying the chan-
nels (red, green, or blue) by a specific factor. In this test, the
color channel is multiplied by the factor (m), where (m = 1..
8); the results are shown in Fig. 6.
We can conclude from Fig. 6 that factor three almost
gives a reasonable gray image; therefore, factor three is a
default color channel, and the user can change it according
to his application and processing. Increasing the value of the
multiplying factor can lead to the brightening of the image;
this can be useful in some cases to extract the object in the
image.
The other test was to test the effect of the divisor on the
result, as shown in Fig. 7, for many images with different
values of the divisor. This test was implemented on color
images by changing the weight of the one-color channel. For
example, in the first row, only the red channel was multiplied
by five, whereas in the second row, the green channel was
FIGURE 7. Effect of dividing image values by a specific number on the
multiplied by five, and in the last row, only the blue channel result of the gray image, where each channel multiplies by five.
was multiplied by five. Increasing the divisor parameters
increases the brightness of the image.
Finally, the proposed method was compared with similar Regarding visual quality, the results produced by the pro-
techniques for converting color images into grayscale images, posed method are significantly superior to those of other
as shown in Fig. 8. techniques.

54634 VOLUME 11, 2023

Z. N. Khudhair et al.: Color to Grayscale Image Conversion Based on SVD

FIGURE 8. Comparing the proposed method with other methods. 1. Color image. 2. Matlab. 3. Reference [26]. 4. Reference [27]. 5. Reference [28]. 6.
Reference [29]. 7. Reference [30]. 8. Reference [6]. 9. Proposed method.

Finally, the quality of the converted images for the dif- IX. CONCLUSION
ferent methods was compared and is listed in Table 1. This paper presents a novel and innovative framework for
It is clear from Table 1 that the proposed method pro- grayscale image conversion from color images. It is a mean-
vides a better image quality than the other methods. The ingful conversion that enhances the usual workflow for ana-
SVD method produces grayscale images with a higher visual lyzing images that work on grayscale images. SVD is the
quality than traditional grayscale conversion methods. This best way to convert a vector into other vectors based on
is because SVD considers the relative importance of each the relation between the input vector values. Furthermore,
channel in an image, resulting in a more accurate grayscale SVD can preserve more color information in an image than
image. other gray-scale conversion methods. This is because SVD

VOLUME 11, 2023 54635

 Z. N. Khudhair et al.: Color to Grayscale Image Conversion Based on SVD

TABLE 1. Compares the gray image quality of various methods.

54636 VOLUME 11, 2023

Z. N. Khudhair et al.: Color to Grayscale Image Conversion Based on SVD

considers the relative importance of each channel in the image [16] S. Liu and X. Zhang, ‘‘Image decolorization combining local fea-
and produces a grayscale image that more accurately captures tures and exposure features,’’ IEEE Trans. Multimedia, vol. 21, no. 10,
pp. 2461–2472, Oct. 2019.
the true intensity values of the image. [17] C. Ancuti, C. O. Ancuti, M. Feixas, and M. Sbert, ‘‘Image decolorization
Unlike traditional methods, the proposed method has high based on information theory,’’ in Proc. IEEE Int. Conf. Image Process.
flexibility for producing more than one version of a grayscale (ICIP), Sep. 2019, pp. 3242–3246.
[18] D. C. Lay, R. L. Steven, and J. McDonald, Linear Algebra and Its Appli-
image. The parameters used in this method can help users cations, 5th ed. Boston, MA, USA: Pearson Education Press, 2016.
produce appropriate images for processing and application. [19] H. Zhang, C. Wang, and X. Zhou, ‘‘A robust image watermarking scheme
based on SVD in the spatial domain,’’ Future Internet, vol. 9, no. 3, p. 45,
In addition, we resolved the problem of producing the same Aug. 2017.
value for vectors of the same magnitude using the weight, [20] K. Santosh and B. Gawali, ‘‘Recent trends in image processing and pat-
which yields different values and grayscale images. This tern recognition,’’ in Proc. 3rd Int. Conf. (RTIPR), Aurangabad, India,
vol. 1380, 2020, pp. 1–213.
method’s grayscale image result is highly promising and has [21] B. Barnes, I. Adjei, C. Sebil, and E. Harris, ‘‘Product-normed linear space,’’
very good quality. This method is more efficient than the other Eur. J. Pure Appl. Math., vol. 11, pp. 740–750, Jul. 2018.
[22] K. J. Audu, ‘‘Application of singular value decomposition technique for
methods. compressing images,’’ Gadau J. Pure Allied Sci., vol. 1, no. 2, pp. 82–94,
Aug. 2022.
ACKNOWLEDGMENT [23] D. Varga, ‘‘No-reference image quality assessment with convolutional
This research was funded by Princess Nourah bint Abdulrah- neural networks and decision fusion,’’ Appl. Sci., vol. 12, no. 101, pp. 1–17,
man University and Researchers Supporting Project number 2022.
[24] A. Rubel, O. Ieremeiev, V. Lukin, J. Fastowicz, and K. Okarma, ‘‘Com-
(PNURSP2023R346), Princess Nourah bint Abdulrahman bined no-reference image quality metrics for visual quality assessment
University, Riyadh, Saudi Arabia. The authors would also like optimized for remote sensing images,’’ Appl. Sci., vol. 12, no. 4, p. 1986,
to thank Prince Sultan University, Riyadh Saudi Arabia for Feb. 2022.
[25] N. Venkatanath, D. Praneeth, M. Chandrasekhar, S. S. Channappayya, and
their support. S. S. Medasani, ‘‘Blind image quality evaluation using perception based
features,’’ in Proc. 21st Nat. Conf. Commun. (NCC), Feb. 2015, pp. 1–6.
[26] A. A. Gooch, S. C. Olsen, J. Tumblin, and B. Gooch, ‘‘Color2Gray:
REFERENCES Salience-preserving color removal,’’ ACM Trans. Graph., vol. 24, no. 3,
[1] Y. Yang, M. Song, J. Bu, C. Chen, and C. Jin, ‘‘Color to gray: Atten- pp. 634–639, 2005.
tion preservation,’’ in Proc. 4th Pacific-Rim Symp. Image Video Technol., [27] M. Grundland and N. A. Dodgson, ‘‘Decolorize: Fast, contrast enhanc-
Nov. 2010, pp. 337–342. ing, color to grayscale conversion,’’ Pattern Recognit., vol. 40, no. 11,
[2] A. Güneş, H. Kalkan, and E. Durmuş, ‘‘Optimizing the color-to-grayscale pp. 2891–2896, Nov. 2007.
conversion for image classification,’’ Signal, Image Video Process., vol. 10, [28] K. Smith, P.-E. Landes, J. Thollot, and K. Myszkowski, ‘‘Apparent
no. 5, pp. 853–860, Jul. 2016. greyscale: A simple and fast conversion to perceptually accurate images
[3] Y. Wan and Q. Xie, ‘‘A novel framework for optimal RGB to grayscale and video,’’ Comput. Graph. Forum, vol. 27, no. 2, pp. 193–200, Apr. 2008.
image conversion,’’ in Proc. 8th Int. Conf. Intell. Hum.-Mach. Syst. Cybern. [29] Y. Kim, C. Jang, J. Demouth, and S. Lee, ‘‘Robust color-to-gray via
(IHMSC), vol. 2, Aug. 2016, pp. 345–348. nonlinear global mapping,’’ ACM Trans. Graph., vol. 28, no. 5, pp. 1–4,
[4] K. Kumar, R. K. Mishra, and D. Nandan, ‘‘Efficient hardware of RGB Dec. 2009.
to gray conversion realized on FPGA and ASIC,’’ Procedia Comput. Sci., [30] C. Lu, L. Xu, and J. Jia, ‘‘Contrast preserving decolorization with
vol. 171, pp. 2008–2015, Dec. 2020. perception-based quality metrics,’’ Int. J. Comput. Vis., vol. 110, no. 2,
[5] M. S. M. Rahim, A. Norouzi, A. Rehman, and T. Saba, ‘‘3D bones pp. 222–239, Nov. 2014.
segmentation based on CT images visualization,’’ Biomed. Res., vol. 28,
no. 8, pp. 3641–3644, 2017.
[6] C. T. Nguyen and J. P. Havlicek, ‘‘Color to grayscale image conversion
using modulation domain quadratic programming,’’ in Proc. IEEE Int.
Conf. Image Process. (ICIP), Sep. 2015, pp. 4580–4584. ZAID NIDHAL KHUDHAIR received the B.Sc.
[7] X. Zhang and S. Liu, ‘‘Contrast preserving image decolorization combin- degree in computer science from the University of
ing global features and local semantic features,’’ Vis. Comput., vol. 34, Kufa, Iraq, in 2010, and the M.Sc. degree from the
nos. 6–8, pp. 1099–1108, Jun. 2018. Faculty of Computing, University of Technology
[8] V. Sowmya, D. Govind, and K. P. Soman, ‘‘Significance of incorporating
Malaysia, in 2013, where he is currently pursuing
chrominance information for effective color-to-grayscale image conver-
sion,’’ Signal, Image Video Process., vol. 11, no. 1, pp. 129–136, Jan. 2017.
the Ph.D. degree. He is an Assistant Lecturer with
[9] T. Saba, S. U. Khan, N. Islam, N. Abbas, A. Rehman, N. Javaid, and the Department of Computer Engineering, Imam
A. Anjum, ‘‘Cloud-based decision support system for the detection and Ja’afar Al-Sadiq University. His research interests
classification of malignant cells in breast cancer using breast cytology include computer vision, digital image processing,
images,’’ Microsc. Res. Technique, vol. 82, no. 6, pp. 775–785, Jun. 2019. medical image processing, data and image secu-
[10] S. Liu, ‘‘Two decades of colorization and decolorization for images and rity, computer networking, and real-time applications.
videos,’’ 2022, arXiv:2204.13322.
[11] C. Kanan and G. W. Cottrell, ‘‘Color-to-grayscale: Does the method
matter in image recognition?’’ PLoS ONE, vol. 7, no. 1, Jan. 2012,
Art. no. e29740.
[12] W. H. Lim and N. A. M. Isa, ‘‘Color to grayscale conversion based on AHMED NIDHAL KHDIAR received the B.Sc.
neighborhood pixels effect approach for digital image,’’ in Proc. of 7th Int. and M.Sc. degrees in electronic and communi-
Conf. Elect. Electron. Eng., Bursa, Turkey, 2011, pp. 1–4. cation engineering from Al-Nahrain University,
[13] K. Padmavathi and K. Thangadurai, ‘‘Implementation of RGB and Iraq, and the Ph.D. degree in electronic engi-
grayscale images in plant leaves disease detection–comparative study,’’
neering/signal processing from USM University.
Indian J. Sci. Technol., vol. 9, no. 6, pp. 1–6, Feb. 2016.
[14] H. Z. Nafchi, A. Shahkolaei, R. Hedjam, and M. Cheriet, ‘‘CorrC2G: Color He was an Academic Staff with Al-Sadiq Uni-
to gray conversion by correlation,’’ IEEE Signal Process. Lett., vol. 24, versity, Islamic University, Al-Nahrain University,
no. 11, pp. 1651–1655, Nov. 2017. and Kufa University, for more than 15 years. He is
[15] N. Damodaran, V. Sowmya, D. Govind, and K. P. Soman, ‘‘Effect of decol- currently an Academic Lecturer and a Researcher.
orized images in scene classification using deep convolution features,’’ He published several articles in scientific journals
Procedia Comput. Sci., vol. 143, pp. 954–961, 2018. and IEEE conferences.

VOLUME 11, 2023 54637

 Z. N. Khudhair et al.: Color to Grayscale Image Conversion Based on SVD

NIDHAL K. EL ABBADI received the B.Sc. TANZILA SABA received the Ph.D. degree in document information secu-
degree in chemical engineering from the Univer- rity and management from the Faculty of Computing, Universiti Teknologi
sity of Baghdad, in 1976, the B.Sc. degree (Hons.) Malaysia (UTM), Malaysia, in 2012. She is currently the Associate Chair
in computer science from Al-Mustansiriya Uni- with the Artificial Intelligence and Data Analytics Laboratory (AIDA),
versity, in 2000, and the M.Sc. and Ph.D. degrees Information Systems Department, College of Computer and Information Sci-
in computer science from the Informatics Institute ence, Prince Sultan University, Riyadh, Saudi Arabia. Her research interests
for Postgraduate Studies (IIPS), Iraq Commission include medical imaging, pattern recognition, data mining, MRI analysis,
for Computer and Informatics, in 2001 and 2006, and soft computing.
respectively. He was an Engineer in many indus-
trial plants. In 2006, he joined Al-Mustansiriya
University, Iraq, as an Academic Staff. He is currently a Professor of
computer techniques engineering with the Al-Mustaqbal University. He has
successfully supervised many master’s and Ph.D. students. He published FATEN S. ALAMRI received the Ph.D. degree in system modeling and
more than 300 articles in various international journals and conference analysis in statistics from Virginia Commonwealth University, USA, in 2020.
proceedings and eight books related to computer science. His research Her Ph.D. research was in Bayesian dose response modeling, experimental
interests include image processing, computer vision, steganography, and the design, and nonparametric modeling. She is currently an Assistant Professor
IoT [ACM Classification Codes (I3, I4, I5)]. He acts as a reviewer for various with the Department of Mathematical Sciences, College of Science, Princess
journals and conferences. He served as the conference chair and a technical Nourah bint Abdulrahman University. Her research interests include spatial
committee member for numerous international conferences. areas, environmental statistics, and brain imaging.

FARHAN MOHAMED (Senior Member, IEEE)

received the B.Sc. and M.Sc. degrees from Uni- AMJAD REHMAN (Senior Member, IEEE)
versiti Teknologi Malaysia (UTM) and the Ph.D. received the Ph.D. and Postdoctoral Research
degree in computer science from Swansea Uni- degrees (Hons.) from the Faculty of Comput-
versity, in 2014. He is currently a Senior Lec- ing, Universiti Teknologi Malaysia, Johor Bahru,
turer with the School of Computing, Faculty of Malaysia, with a focus on forensic documents
Engineering, UTM. He is also a Research Fellow analysis and security, in 2010 and 2011, respec-
with the Media and Games Innovation Centre of tively. He is currently a Senior Researcher with
Excellence (MaGICX), Institute of Human Cen- the Artificial Intelligence and Data Analytics Lab-
tred Engineering, UTM. He led a study with SME oratory (AIDA), College of Computer and Infor-
Corp-Huawei in analyzing SME digitalization, in 2018. His current works mation Science, Prince Sultan University, Riyadh,
emphasize visual analytics, virtual environment, and procedural computer Saudi Arabia. He is a PI in several funded projects and also completed
graphics. He received the Leaders in Innovation Fellowship from Newton projects funded by MOHE Malaysia, Saudi Arabia. His research interests
Fund, Royal Academy of Engineering, U.K., in 2017. He is on the Executive include data mining, health informatics, and pattern recognition.
Committee of the IEEE Computer Society Malaysia.

54638 VOLUME 11, 2023