Array 6 (2020) 100024

Contents lists available at ScienceDirect

Array
journal homepage: www.elsevier.com/journals/array/2590-0056/open-access-journal

Image compression based on 2D Discrete Fourier Transform and matrix

minimization algorithm
Mohammed H. Rasheed a, Omar M. Salih a, Mohammed M. Siddeq a, *, Marcos A. Rodrigues b
a
Computer Engineering Dept., Technical College/Kirkuk, Northern Technical University, Iraq
b
Geometric Modeling and Pattern Recognition Research Group, Sheffield Hallam University, Sheffield, UK

A R T I C L E I N F O A B S T R A C T

Keywords: In the present era of the internet and multimedia, image compression techniques are essential to improve image
DFT and video performance in terms of storage space, network bandwidth usage, and secure transmission. A number of
Matrix minimization algorithm image compression methods are available with largely differing compression ratios and coding complexity. In this
Sequential search algorithm
paper we propose a new method for compressing high-resolution images based on the Discrete Fourier Transform
(DFT) and Matrix Minimization (MM) algorithm. The method consists of transforming an image by DFT yielding
the real and imaginary components. A quantization process is applied to both components independently aiming
at increasing the number of high frequency coefficients. The real component matrix is separated into Low Fre-
quency Coefficients (LFC) and High Frequency Coefficients (HFC). Finally, the MM algorithm followed by
arithmetic coding is applied to the LFC and HFC matrices. The decompression algorithm decodes the data in
reverse order. A sequential search algorithm is used to decode the data from the MM matrix. Thereafter, all
decoded LFC and HFC values are combined into one matrix followed by the inverse DFT. Results demonstrate that
the proposed method yields high compression ratios over 98% for structured light images with good image
reconstruction. Moreover, it is shown that the proposed method compares favorably with the JPEG technique
based on compression ratios and image quality.

1. Introduction use [2,3]. Basically, if the purpose of an image is to be seen by humans

then we can assume that the image can have a variable high level of
The exchange of uncompressed digital images requires considerable redundant data. Redundant data in digital images come from the fact that
amounts of storage space and network bandwidth. Demands for efficient pixels in digital images are highly correlated to a level where reducing
image compression result from the widespread use of the Internet and this correlation cannot be noticed by the human eye (Human Visual
data sharing enabled by recent advances in digital imaging and multi- System) [4,5]. Consequently, most of these redundant, highly correlated
media services. Users are creating and sharing images with increased size pixels can be removed while maintaining an acceptable level of human
and quantity and expect quality image reconstruction. It is clear that visual quality of the image. Therefore, in digital images the Low Fre-
sharing multimedia-based platforms such as Facebook and Instagram quency Components (LFC) are more important as they contribute more to
lead to widespread exchange of digital images over the Internet [1]. This define the image contents than High Frequency Components (HFC).
has led to efforts to improve and fine-tune present compression algo- Based on this, the intension is to preserve the low frequency values and
rithms along with new algorithms proposed by the research community shorten the high frequency values by a certain amount, in order to
to reduce image size whilst maintaining the best level of quality. For any maintain the best quality with the lowest possible size [6,7].
digital image, it can be assumed that the image in question may have Image frequencies can be determined through a number of trans-
redundant data and can be neglected to a certain extent. The amount of formations such as the Discrete Cosine Transform (DCT), Discrete
redundancy is not fixed, but it is an assumed quantity and its amount Wavelet Transform (DWT) and Discrete Fourier Transform (DFT) [8]. In
depends on many factors including the requirements of the application to this study we will use DFT as a first step in the process to serialize a digital
be used, the observer (viewer) or user of the image and the purpose of its image for compression. Since its discovery, the DFT has been used in the

* Corresponding author.
E-mail addresses: [email protected] (M.H. Rasheed), [email protected] (O.M. Salih), [email protected] (M.M. Siddeq), M.Rodrigues@shu.
ac.uk (M.A. Rodrigues).

https://doi.org/10.1016/j.array.2020.100024
Received 21 November 2019; Received in revised form 11 February 2020; Accepted 6 March 2020
Available online 8 March 2020
2590-0056/© 2020 The Authors. Published by Elsevier Inc. This is an open access article under the CC BY license (http://creativecommons.org/licenses/by/4.0/).
M.H. Rasheed et al. Array 6 (2020) 100024

Fig. 1. The proposed compression method.

Fig. 2. DFT applied to a 4
4 matrix of data.

Fig. 3. Quantization and rounding off the real and imaginary components.

Ref. [10,11]. The main purpose of matrix minimization is to reduce High

Frequency Components (HFC) to 1/3 of its original size by converting
each three items of data into one, a process that also increases redundant
coefficients [11,12]. The main problem with Matrix Minimization is that
it has a large probability data called Limited-Data [13,14,16]. Such
probabilities are combined within the compressed file as indices used
Fig. 4. Each block (4
4) is divided to real and imaginary matrices (after later in decompression.
applying DFT). The real matrix contains DC value at first location, these DC Our previous research [13,14] used the DCT combined with Matrix
values are saved in a new matrix. The rest of high-frequency coefficients are Minimization algorithm yielding over 98% compression ratios for
saved in a different matrix as shown in contents of the LFC-Matrix, HFCReal structured light images and 95% for conventional images. The main
and HFCImag. justification to use DFT in the proposed method is to demonstrate that the
Matrix Minimization algorithm is very effective in connection with a
field of image processing and compression. The DFT is used to convert an discrete transform and, additionally, to investigate the DFT for image
image from the spatial domain into frequency domain, in other words it compression.
allows us to separate high frequency from low frequency coefficients and The contribution of this research is to reduce the relatively large
neglect or alter specific frequencies leading to an image with less infor- probability table to two values only, minimum and maximum, rather
mation but still with a convenient level of quality [8–10]. than keeping the entire lookup table (referred to as Limited-Data in our
We propose a new algorithm to compress digital images based on the previous research [10,11,12and13]). The main reason is to increase
DFT in conjunction with the Matrix Minimization method as proposed in compression ratios by reducing the size of the compressed file header.

2
M.H. Rasheed et al. Array 6 (2020) 100024

Fig. 5. The Matrix Minimization method for an m x n matrix [10–12].

block representing the real and the imaginary parts respectively.

Regarding the real part, all low coefficient values (i.e. the DC values) are
detached and saved into a new matrix called Low Frequency Coefficients
(LFC-Matrix) and its substituted with a zero value in the quantized ma-
trix. It is important to note that DC values are only found in the real parts
which highly contribute to the main details and characteristics of the
image. The generated LFC-Matrix size consists of all the DC values of the
entire image can be considered small compared to all other High Fre-
quency Coefficients (HFC-Matrix) and can be represented with few bytes.
Fig. 6. Separating zeros and nonzero from HFC matrix and coding zero and non-
Fig. 4 illustrates the content of the generated three matrices.
zero values into Zero and Value matrices.
Since the size of the LFC-Matrix is small compared to HFC-Matrices, it
is very obvious that HFC matrices for both real and imaginary parts need
The proposed compression algorithm is evaluated and analyzed through to be reduced to get a reasonable compression. Therefore, the algorithm
measures of compression ratios, RMSE (Root Mean Square Error) and called Matrix-Minimization suggested by Siddeq and Rodrigues [10] is
PSNR (Peak Signal-to-Noise Ratio). It is demonstrated that the proposed applied. The algorithm is used to reduce the size of HFC matrices by
method compares well with the popular JPEG technique. contracting every three coefficients to a single equivalent value, which
can be traced back to their original values in the decompression phase.
2. The proposed compression algorithm The contraction is performed on each three consecutive coefficients using
Random-Weight-Values. Each value is multiplied by a different random
The proposed compression method is illustrated in Fig. 1. Initially, an number (Ki) and then their summation is found, the value generated is
original image is subdivided into non-overlapping blocks of size M x N considered a contracted value of the input values. Fig. 5 illustrates the
pixels starting at the top left corner of the image. The Discrete Fourier Matrix Minimization applied to M x N matrix [11,12].
transform (DFT) is applied to each M x N block independently to repre- It is important to note that in the decompression phase a search al-
sent the image in the frequency domain yielding the real and imaginary gorithm is required to find the three original values that are used to find
components. The Matrix Minimization algorithm is applied to each the contracted value, therefore, the minimum and maximum values of
component and zeros are removed. The resulting vectors are subjected to the m x n block are stored. The idea behind this is to limit the range of
Arithmetic coding and represent the compressed data. values required to recover the original three values that made the con-
To illustrate the process for each M x N (M ¼ N ¼ 4) block in the tracted value hence increase the speed of the search algorithm at
original image, we represent a 4
4 block in Fig. 2 below: decompression stage.
A uniform quantization is then applied to both parts, which involves Because in previous work the range of the search space are limited in
dividing each element by a factor called quantization factor Q followed the array for easy searching and this was encoded in the header file to be
by rounding the outcomes which results in an increase of high frequency used at decompression stage. However, it is possible that complex images
coefficients probability thus reducing the number of bits needed to may generate large arrays which, in turn, will impair compression (make
represent such coefficients. The result of this operation is that the it more computationally demanding). For this reason, we suggested
compression ratio increases. Fig. 3 illustrate the quantization and another method in this paper using DFT and reduced limited search area
rounding off steps. For more information, the uniform quantization (Qr (i.e. search area contains just two values [MIN, MAX]). Such bounding
and Qi) are selected heuristically. makes searching for the sought values easier and faster. Any further
Up to this point, two matrices (Qr and Qi) have been generated per detailed information about Matrix Minimization can be found in

Fig. 7. Decompression steps.

3
M.H. Rasheed et al. Array 6 (2020) 100024

Fig. 8. (a), (b) and (c) Lena, Lion and Apple images status are compressed by our proposed method using different quantization values.

references [11,12,16]. These three references show with examples how zero value in the HFC-Matrix is counted during the process. A new
the Matrix Minimization works with keys and how the limited search is array called Zero Matrix is then created in which we append a zero value
used for decoding. whenever we have a non-zero value at the same index in the original
After the Matrix-Minimization algorithm has been applied, the pro- HFC-Matrix followed by an integer that represents the total number of
duced HFC-Matrix for both real and imaginary parts are examined and it zeros between any two non-zero values. Fig. 6 demonstrates the process
is possible to see a high probability in the number of zero values than any of separating zeros and non-zero values [14–16].
other values in the matrix. Therefore, separating zero from non-zero The zero values in the Zero-Matrix reflect the actual non-zero values
values will remove redundant data and hence increase the efficiency of in sequences in the original matrix. Likewise, the integer values reflect
the arithmetic coding compression [9,10,13,14]. the total number of zeros that come thereafter. Finally, the two matrices
The implementation of the method is by isolating all zero values from are ready for compression by a coding method which in our case is
the matrix while preserving all non-zero values in a new array called arithmetic coding [6,7]. It is important to note that the proposed method
Value Matrix. The total number of zeros removed between each non- described above is also applied to the LFC-Matrix which contains the low

4
M.H. Rasheed et al. Array 6 (2020) 100024

Fig. 9. (a), (b) and (c) Boeing 777, Girl and Baghdad colour images are compressed by our proposed method using different quantization values.

frequency coefficients values of the real part. Up to this point, the called Sequential Search Algorithm [10–13], which is based on three
Value-Matrix and Zero-Matrix in our case are considered headers and pointers working sequentially to regenerate the three values that
used in the decompression process to regenerate the original HFC and constitute the contracted values with assistance of the MIN and MAX
LFC matrices. values which were preserved during the compression process. The MIN
and MAX values are considered to be the limited space search values used
3. The decompression algorithm to restore the actual HFC for both parts (real and imaginary) [14–18].
Finally, an inverse quantization and DFT is applied to each part to
The decompression algorithm is a counter compression operation reconstruct the compressed digital image. Fig. 7 illustrates the decom-
which performs all functions of the compression but in reverse order. The pression steps.
steps to decompression start by decoding the LFC-Matrix, Value-Matrix
and Zero-Matrix using arithmetic decoding followed by reconstructing a 4. Experimental results
unified array based on Value and Zero matrices and reconstruct the HFC-
Matrix for both parts. Siddeq and Rodrigues proposed a novel algorithm Experimental results shown here demonstrate the efficacy of the

5
M.H. Rasheed et al. Array 6 (2020) 100024

Fig. 10. (a) The 3D Scanner developed by the GMPR group, (b) a 2D picture captured by the camera, (c) 2D image converted into a 3D surface patch.

Fig. 11. Original 2D images with different dimensions used by our proposed compression method.

 We apply the proposed compression technique to structured light

Table 1
images (i.e. a type of image used for reconstruct 3D surfaces - see
Results for grey images.
Section 5).
Image Image Quantization After (Bit/ RMSE PSNR
Size Compression Pixel)
5. Results for structured light images and 3D surfaces
(MB) (KB) bpp

Lena 1.0 10 260 0.253 1.2 47.3

A 3D surface mesh reconstruction method was developed by Rodri-
25 138.2 0.134 2.4 44.3
45 88.1 0.086 3.9 42.2
gues [8,19] with a team within the GMPR group at Sheffield Hallam
Lion 1.37 25 201 0.143 2.5 44.1 University. The working principle of the 3D mesh scanner is that the
60 108.4 0.077 5.0 41.1 scene is illuminated with a stripe pattern whose 2D image is then
100 71.4 0.05 8.1 39.0 captured by a camera. The relationship between the light source and the
Apples 1.37 10 228 0.162 1.2 47.3
camera determines the 3D position of the surface along the stripe pattern.
30 91.7 0.065 2.6 43.9
60 47.8 0.034 4.6 41.5 The scanner converts a surface to a 3D mesh in a few milliseconds by

proposed compression technique. Our proposed method was imple- Table 3

mented in MATLAB R2014a running on an Intel Core i7-3740QM Compressed 2D structured light images.
microprocessor (8-CPUs). For clarity, we divide the results into two parts: Image Image Quantization After (bit/ RMSE PSNR
Size Compression Pixel)
 The method applied to general 2D images of different sizes and assess (MB) (KB) bpp
their visual quality with RMSE [1,3]. Also, we applied Peak Corner 1.25 60 35.4 0.027 4.7 41.4
Signal-to-Noise Ratio (PSNR) for measuring image quality. This 100 17.8 0.013 15.5 36.2
measurement widely used in digital image processing [23]. Tables 1 Face1 1.37 100 34.0 0.024 8.4 38.8
160 18.1 0.012 11.5 37.5
and 2 show the first part of results by applying the proposed com-
Face2 1.37 50 46.2 0.032 6.7 39.8
pression/decompression method to six selected images whose details 150 20.1 0.014 9.9 38.1
are shown in Figs. 8 and 9.

Table 2
Results for colour images.
Image Image Size (MB) Quantization for each layer in R,G,B After Compression (KB) (Bit/Pixel) bpp RMSE PSNR

Boeing 777 6.15 10 437.4 0.069 2.1 44.9

25 182.8 0.029 3.9 42.2
Girl 4.29 10 641.1 0.145 3.9 42.2
25 315.6 0.071 5.5 40.7
Baghdad 8.58 25 426.3 0.097 4.4 41.6
35 309.8 0.07 5.6 40.6

6
M.H. Rasheed et al. Array 6 (2020) 100024

Fig. 12. (a) and (b): shows the 2D decompressed for Corner’s image, that used in 3D application to reconstruct 3D mesh surface. The 3D mesh (3D vertices and
triangles) is successfully reconstructed without significant distortion at high compression ratios up to 98.6%.

7
M.H. Rasheed et al. Array 6 (2020) 100024

Fig. 13. (a) and (b): shows decompressed for Face1 2D image, that used in the 3D application to reconstruct 3D mesh surface. The 3D mesh is successfully recon-
structed without significant distortion at high compression ratios of 98.6%.

8
M.H. Rasheed et al. Array 6 (2020) 100024

Fig. 14. (a) and (b): shows decompressed for Face2 2D image, that used in 3D application to reconstruct 3D mesh surface. The 3D mesh was successfully reconstructed
without significant distortion at high compression ratios of 98.5%.

9
M.H. Rasheed et al. Array 6 (2020) 100024

Table 4
Comparative analysis of using DFT alone and our proposed method (DFT and Matrix Minimization) based on image quality and compressed size.
Image Size (MB) Quantization Factor Compressed size DFT alone RMSE PSNR Compressed Size DFT þ MM RMSE PSNR

KB bpp KB bpp

Conventional images
Lena 1.0 45 721 0.7 2.1 44.9 88 0.085 3.9 42.2
Lion 1.37 100 808 0.57 3.59 42.5 71 0.05 8.1 39.0
Apples 1.37 60 576 0.41 2.3 44.5 47 0.033 4.6 41.5
Boeing 6.15 25 2200 0.35 1.8 45.5 182 0.028 3.9 42.2
Girl 4.29 25 2350 0.54 3.59 42.5 315 0.071 5.5 40.7
Bagdad 8.58 35 3500 0.4 2.3 44.5 309 0.035 5.6 40.6
Structured light images
Corner 1.25 100 615 0.48 2.9 43.5 17 0.013 15.5 36.2
Face1 1.37 160 624 0.44 4.8 41.3 18 0.012 11.5 37.5
Face2 1.37 150 508 0.36 4.1 42.0 20 0.014 9.9 38.1

Table 5
Comparative analysis of compression using JPEG and our approach based on image quality and compression size.
Image Size (MB) Compressed Size by JPEG RMSE PSNR Compressed Size by DFT þ MM RMSE PSNR

KB bpp KB bpp

Conventional images
Lena 1.0 64 0.062 1.9 45.3 88 0.085 3.9 42.2
Lion 1.37 56 0.039 8.8 38.6 71 0.05 8.1 39.0
Apples 1.37 48 0.034 3.2 43.0 47 0.033 4.6 41.5
Boeing 6.15 210 0.033 8.7 38.7 182 0.028 3.9 42.2
Girl 4.29 347 0.078 9.8 38.2 315 0.071 5.5 40.7
Bagdad 8.58 279 0.031 3.5 42.6 309 0.035 5.6 40.6
Structured light images
Corner 1.25 26 0.02 14.3 36.5 17 0.013 15.5 36.2
Face1 1.37 23 0.016 16.5 35.9 18 0.012 11.5 37.5
Face2 1.37 27 0.019 13.1 36.9 20 0.014 9.9 38.1

Fig. 15. Compressed and decompressed greyscale images by JPEG, the quality of the decompressed images varies compared with our approach according to RMSE
and PSNR.

using a single 2D image [19,20] as shown in Fig. 10. method described in this paper whose compressed sizes with RMSE and
The significance of using such 2D images is that, if the compression PSNR are shown in Table 3. After decompression, the images were sub-
method is lossy and results in a noisy image, the 3D algorithms will jected to 3D reconstruction using the GMPR method and compared with
reconstruct the surface with very noticeable artefacts, that is, the 3D 3D reconstruction of the original images. The reconstructed 3D surfaces
surface becomes defective and degraded with problem areas easily are shown in Figs. 12–14.
noticeable. If, on the other hand, the 2D compression/decompression is
of good quality, then the 3D surface is reconstructed well and there are no 6. Discussion and comparative analysis
visible differences between the original reconstruction and the recon-
struction with the decompressed images. Our literature survey did not show results for image compression
Fig. 10 (left) depicts the GMPR scanner together with an image using the DFT alone. The reason is that by applying a DFT, it yields two
captured by the camera (middle) which is then converted into a 3D sets of coefficients, real and imaginary. If one wishes to keep those for
surface and visualized (right). Note that only the portions of the image faithful image reconstruction, then it is not possible to achieve high
that contain patterns (stripes) can be converted into 3D; other parts of the compression ratios. We applied the DFT as described in this paper
image are ignored by the 3D reconstruction algorithms [21,22]. The resulting in images with good visual quality and low compression
original images used in this research are shown in Fig. 11 (Corner, Face1 complexity. A comparative analysis between compression ratios for DFT
and Face2). The three images shown in Fig. 11 were compressed by the alone and DFT followed by the Matrix Minimization algorithm show

10
M.H. Rasheed et al. Array 6 (2020) 100024

Fig. 16. Compressed and Decompressed colour images by JPEG, the decompressed images (Boeing and Girl) have lower quality compared with our approach ac-
cording to RMSE and PSNR. However, our approach couldn’t reach to JPEG level of compression for Bagdad’s image on the right.

Fig. 17. 3D reconstruction from JPEG compressed images. In (a) reconstruction was possible but with significant artefacts. In (b) 3D reconstruction was not possible as
images were too deteriorated.

Table 6
Comparative analysis between pervious work [12] (Matrix Minimization algorithm) and our approach based on time execution.
Image Size (MB) Previous work (Matrix Minimization algorithm) The proposed algorithm

Compressed size (KB) Bits/Pixel (bpp) Decompression time (seconds) Compressed size (KB) Bits/pixel (bpp) Decompression time (seconds)

Lena 1.0 120 0.117 102 88 0.085 25

Lion 1.37 98 0.069 240 71 0.05 40
Apples 1.37 92 0.065 90 47 0.033 15
Boeing 6.15 240 0.038 420 182 0.028 150
Girl 4.29 399 0.090 330 315 0.071 114
Bagdad 8.58 673 0.076 720 309 0.035 198
Corner 1.25 56 0.043 84 17 0.013 22
Face1 1.37 46 0.032 144 18 0.012 59
Face2 1.37 38 0.027 174 20 0.014 66

11
M.H. Rasheed et al. Array 6 (2020) 100024

Table 7
Comparative analysis between pervious work [12] (Matrix Minimization) and our approach based on image quality and compression sizes.
Image Size (MB) Previous work (Matrix Minimization algorithm) The proposed algorithm

Compressed Size (KB) Bits/Pixel (bpp) RMSE PSNR Compressed Size (KB) Bits/Pixel (bpp) RMSE PSNR

Lena 1.0 120 0.117 6.8 39.8 88 0.085 3.9 42.2

Lion 1.37 98 0.069 10.1 38.0 71 0.050 8.1 39.0
Apples 1.37 92 0.065 7.1 39.6 47 0.033 4.6 41.5
Boeing 6.15 240 0.038 10.2 38.0 182 0.028 3.9 42.2
Girl 4.29 399 0.090 8.4 38.8 315 0.071 5.5 40.7
Bagdad 8.58 673 0.076 5.9 40.4 309 0.035 5.6 40.6
Corner 1.25 56 0.043 16.0 36.0 17 0.013 15.5 36.2
Face1 1.37 46 0.032 14.4 36.5 18 0.012 11.5 37.5
Face2 1.37 38 0.027 11.2 37.6 20 0.014 9.9 38.1

enormous differences as shown in Table 4. 7. Conclusion

The results demonstrate that our proposed method of using a DFT in
conjunction with the Matrix Minimization algorithm has the ability to This research has demonstrated a novel approach to compress images
compress digital images up to 98% compression ratios. It is shown that in greyscale, colour and structured light images used in 3D reconstruc-
the DFT alone cannot compress images with similar ratios and quality. tion. The method is based on the DFT and the Matrix-Minimization al-
Although it can be seen from Table 4 that our proposed method (DFT þ gorithm. The most important aspects of the method and their role in
Matrix Minimization algorithm) increases the overall RMSE and, while providing high quality image with high compression ratios are high-
some image details are lost, reconstructed images are still high quality. lighted as follows.
Additionally, the proposed method is compared with JPEG technique
[23–25] which is a popular technique used in image and video  After dividing an image into non-overlapping blocks (4
4), a DFT is
compression. Also, the JPEG is used in many areas of digital image applied to each block followed by quantizing each part (real and
processing [26]. The main reason for comparing our method with JPEG is imaginary) independently. Meanwhile, the DC value (Low Frequency
because JPEG is based on DCT and Huffman coding. Table 5 shows the Coefficients) from each block are stored in a new matrix, while the
analytical comparison between the two methods. rest of the values in the block are the High Frequency Coefficients.
In above Table 5 it shown that our proposed method is better than  The Matrix-Minimization algorithm is applied to reduce the high-
JPEG technique to compress structured light images, while for conven- frequency matrix to 1/3 of its original size, leading to increased
tional images it can be stated that both methods are roughly equivalent as compression ratios.
image quality varies in both methods. The following Figs. 15–17 show  The relatively large probability table of previous method was reduced
comparisons between our approach the and JPEG technique for the im- to two values, minimum and maximum leading to higher compression
ages shown in Tables 4 and 5 ratios and faster reconstruction.
Concerning the compression of structured light images for 3D mesh
reconstruction, the comparison of our method with JPEG shows enor- Results demonstrate that our approach yields better image quality at
mous potential for our approach as depicted in Figs. 11–14. Trying to higher compression ratios while being capable of accurate 3D recon-
compress the same images using JPEG and then using the decompressed struction of structured light images at very high compression ratios.
image to generate the 3D mesh clearly shows the problems and limita- Overall, the algorithm yields a best performance on colour images and
tions of JPEG. This is illustrated in Fig. 17, which shows the JPEG structured light images used in 3D reconstruction than on standard grey
technique on two structured light images for 3D mesh reconstruction. images.
Comparative analysis focused on our previous work on the Matrix On the other hand, the compression steps introduced by the MM al-
Minimization algorithm based on two discrete transforms DWT and DCT, gorithm, especially at decompression stage, make the compression al-
as suggested by Siddeq and Rodrigues [9–11] performing compression gorithm more complex than, for instance, standard JPEG. In general, it
and encryption at the same time. However, complexity of compression can be stated that decompression is slower than compression due to the
and decompression algorithms is cited as a disadvantage of previous search space to recover the original Low and High Frequency coefficients.
work. Table 6 shows the decompression time for the Matrix Minimization In addition, arithmetic coding and decoding is applied to three sets of
algorithm [12] (previous work) compared with our proposed approach. data (DC values, in addition to real and imaginary frequency coefficients)
The advantages of the proposed over previous work are summarized as adding significantly more computation steps leading to increased
follows: execution time.

 The complexity of the decompression steps is reduced in the proposed Conflicts of interest
approach. This is evident from execution times quoted in Table 6 as
the current approach runs faster than previous work on the same The authors declare that there are no conflicts of interest regarding
hardware. the publication of this paper.
 The header file information of current approach is smaller than pre-
vious work leading to increased compression ratios. CRediT authorship contribution statement

It is important to stress the significant novelties of the proposed Mohammed H. Rasheed: Conceptualization, Methodology. Omar
approach which are the reduced number of steps at decompression stage M. Salih: Data curation, Writing - original draft. Mohammed M. Siddeq:
and smaller header information resulting in faster reconstruction from Visualization, Software. Marcos A. Rodrigues: Supervision, Writing -
data compressed at higher compression ratios. Table 7 shows that our review & editing.
proposed image compression method has higher compression ratios and
better image quality (i.e. for both types conventional and structured light Acknowledgments
images) as measured by RMSE and PSNR.
We grateful acknowledge the Computing, Communication and

12
M.H. Rasheed et al. Array 6 (2020) 100024

Cultural Research Institute (C3RI) and the Research and Innovation Of- [13] Siddeq Mohammed, Rodrigues Marcos. A novel high frequency encoding algorithm
for image compression. EURASIP J Appl Signal Process 2017;26. https://doi.org/
fice at Sheffield Hallam University for their support.
10.1186/s13634-017-0461-4.
[14] Siddeq Mohammed, Rodrigues Marcos. DCT and DST based image compression for
References 3D reconstruction. 3D Research 2017;8(5):1–19.
[15] Sheffield Hallam University, Mohammed M Siddeq and Marcos A Rodrigues. Image
[1] Richardson IEG. Video codec design. John Wiley &Sons; 2002. data compression and decompression using minimize size matrix algorithm. WO
[2] Sayood K. Introduction to data compression. 2nd ed. Academic Press, Morgan 2016/135510 A1. Patent 2016.
Kaufman Publishers; 2001. [16] M.M. Siddeq and Rodrigues Marcos. Novel 3D compression methods for geometry,
[3] Rao KR, Yip P. Discrete cosine transform: algorithms, advantages, applications. San connectivity and texture. 3D Research, 7 (13). 2016
Diego, CA: Academic Press; 1990. [17] Siddeq MM, Rodrigues Marcos. 3D point cloud data and triangle Face compression
[4] Gonzalez Rafael C, Woods RichardE. Digital image processing. Addison Wesley by a novel geometry minimization algorithm and comparison with other 3D
publishing company; 2001. formats. In: Proceedings of the international conference on computational methods.
[5] Yuan Shuyun, Hu Jianbo. Research on image compression technology based on vol. 3. California USA: University of California; 2016. p. 379–94.
Huffman coding. J Vis Commun Image Represent February 2019;59:33–8. [18] Siddeq MM, Rodrigues AM. A novel hexa data encoding method for 2D image
[6] Li Peiya, Lo Kwok-Tung. Joint image encryption and compression schemes based on crypto-compression. Multimed Tool Appl 2019. https://doi.org/10.1007/s11042-
16
16 DCT. J Vis Commun Image Represent January 2019;58:12–24. 019-08405-3. Springer.
[7] M. Rodrigues, A. Robinson and A. Osman. Efficient 3D data compression through [19] Rodrigues M, Kormann M, Schuhler C, Tomek P. Robot trajectory planning using
parameterization of free-form surface patches, In: Signal process and multimedia OLP and structured light 3D machine vision. Heidelberg: Springer; 2013. p. 244–53.
applications (SIGMAP), proceedings of the 2010 international conference on. IEEE, Lecture notes in Computer Science Part II. LCNS, 8034 (8034).
130-135. [20] Rodrigues M, Kormann M, Schuhler C, Tomek P. Structured light techniques for 3D
[8] Siddeq MM, Al-Khafaji G. Applied minimize-matrix-size algorithm on the surface reconstruction in robotic tasks. In: Kacprzyk J, editor. Advances in
transformed images by DCT and DWT used for image compression. Int J Comput intelligent systems and computing. Heidelberg: Springer; 2013. p. 805–14.
Appl 2013;70:15. [21] Rodrigues M, Kormann M, Schuhler C, Tomek P. An intelligent real time 3D vision
[9] Siddeq MM, Rodrigues MA. A new 2D image compression technique for 3D surface system for robotic welding tasks. Mechatronics and its applications. IEEE Xplore;
reconstruction. In: 18th international conference on circuits, systems, 2013. p. 1–6.
communications and computers. Greece: Santorin Island; 2014. p. 379–86. [22] Wang, Zhou; Bovik, A.C.; Sheikh, H.R.;Simoncelli, E.P. Image quality assessment:
[10] Siddeq MM, Rodrigues MA. A novel image compression algorithm for high 2004 from error visibility to structural similarity". IEEE Trans Image Process. 13(4):
resolution 3D reconstruction. 3D Research 2014;5(2). https://doi.org/10.1007/ 600–612.
s13319-014-0007-6. Springer. [23] Adler A, Boublil D, Zibulevsky M. Block-based compressed sensing of images via
[11] M.M. Siddeq and Rodrigues Marcos. Applied sequential-search algorithm for deep learning. In: 2017 IEEE 19th international workshop on multimedia signal
compression-encryption of high-resolution structured light 3D data. In: Blashki, processing. Luton: MMSP; 2017. p. 1–6.
Katherine and Xiao, Yingcai, (eds.)MCCSIS : multi conference on computer science [24] BAl-Ani MuzhirShaban, Hammouri Talal Ali. Video compression algorithm based
and information systems 2015. IADIS Press, 195-202. on frame difference approaches. Int J Soft Comput November 2011;2(No.4).
[12] Siddeq MM, Rodrigues Marcos. A novel 2D image compression algorithm based on [25] Adler A. Covariance-assisted matching pursuit. IEEE Signal Process Lett Jan. 2016;
two levels DWT and DCT transforms with enhanced minimize-matrix-size algorithm 23(1):149–53.
for high resolution structured light 3D surface reconstruction. 3D Research 2015; [26] Dar Y, Elad M, Bruckstein AM. Optimized pre-compensating compression. IEEE
6(3):26. https://doi.org/10.1007/s13319-015-0055-6. Trans Image Process Oct. 2018;27(10):4798–809.

13