Noname manuscript No.

(will be inserted by the editor)

Image processing: Image compression using

compressed sensing, discrete cosine transform and
wavelet transform.

[Link] · [Link]

Received: 30 June 2021 / Accepted: Date

Abstract Recently, the image quality and the speed of acquisition are
very studied, particularly in the medical field. We try to find in our
paper the most efficient compression method, which allows having
good image quality with a short compression period. Reducing the
compression time amounts to reducing the acquisition time. In our
article, we proposed three compression methods applied to the
medical image: the discrete cosine transform (DCT) method, the
wavelet transformation (DWT) and the compressed sensing method
(CS). We studied also the acquisition time. In our results, we found that
the DWT method gives better image quality compared to other
methods. We found also that the CS method is faster than other
methods. .
Keywords: Compression image, Discrete Cosine (DCT), Wavelet
Transform (WT), compressed sensing (CS), medical image.

1 Introduction

Image quality is very important in the medical field. It gives more information
to the physical, anatomical or functional data of a human. Good medical image
quality helps radiologists and doctors to give the correct diagnosis. In addition,
the acquisition time is very important in medical imaging .For faster decision

[Link]
Biomedical engineering. University of Abou bekr belkaid faculty of technology
tlemcen,13000 Tlemcen. Algeria .

E-mail: [Link]@[Link]
A. Korti
Biomedical engineering. University of Abou bekr belkaid faculty of technology tlemcen,13000
Tlemcen. Algeria . E-mail: [Link]@[Link]

665
2 [Link], [Link]

about doctors and for the comfort of the patient, it is necessary to reduce
the acquisition time.
Image compression is one of the most widely used in image processing
techniques today. Its role is to reduce the size of the image in order to re-
duce the space. It facilitates processing with a reduced number of data in
a short time. Different compression methods used such as the discrete co-
sine Transform (DCT) (Nasir Ahmed.1972), Wavelet transform (DWT) (Alfred
Haar.1909)[2],JPEG2000 (the Joint photographic Experts Group working
group.1997-2000)[10], etc.)
Unfortunately, the compressed image presents a loss of information, which
degrades the image quality. In this article, we improved the image quality by
using less data. We used compression methods to choose the most relevant
data. We chose three compression methods: DCT, DWT and CS. In the fol-
lowing of this article, we presented compression methods used here and the
different results obtained from the compression methods with discussion. We
ended this article with a conclusion.

2 Methods

In our work, we used three compression methods: DCT, DCT and CS

2.1 Discrete Cousin Transform (DCT)

The Discrete Cosine Transform (DCT) is a type of fast computing Fourier

transform, which maps real signals to corresponding values in frequency do-
main. The DCT method works on the real part of the complex signal because
most of the real-world signals are real signals with no complex components[1].
We will discuss the implementation of DCT algorithm on medical Image Data
The Figure 1 represent the diagram that explains the steps in reconstructed
compressed image using DCT. The information in the frequency domain DCT
(i, j) is obtained from the discrete data of the image img (x, y) where the X
and Y axes are the horizontal and vertical dimensions of the image[3] ,[4],[11].
In this article, we applied the DCT line by line on our image. The compressed
image img’(x, y) is obtained from the frequency data DCT (i, j) by applying
the inverse transform DCT (i,j)-1 .

Fig. 1 Diagram for compressed image using DCT

666
Title Suppressed Due to Excessive Length 3

2.2 Discrete wavelet transform (DWT)

The Discrete Wavelet Transform (DWT) is a representation of signals or images

in a time-frequency form. The Figure 2 represents the diagram, which explains
the reconstruction of the compressed image using wavelets [6], [12]. The time-
frequency data obtained from the original image and the compressed image
reconstructed from the time-frequency data.

Fig. 2 Diagram for compressed image using DWT

2.3 Compressed Sensing (CS)

Compressed sensing (CS) is a newer compression method. It based on three

essential points:
• Sparsity: the desired signal must have a sparse representation in a known
transformation domain.
• Incoherency: The subsampled space should generate aliasing artifacts
similar to noise in the compression transform domain.
• Nonlinear reconstruction: A nonlinear reconstruction is necessary to
exploit the Sparsity while maintaining the consistency of the data acquires
[7],[8] ,[9]To meet the need for the Sparsity, we applied a mask on our
data[5]
The figure 3 represent the diagram, which explains the steps of recon-
structed compressed image using CS.

667
 4 [Link], [Link]

Fig. 3 Diagram for compressed image using CS.

3 Results and Discussions

In our applications, we used the Matlab 2014a language on a dual graphics

card I5 PC. In order to study the performance of the compression methods,
we chose a phantom image of size 512 * 512. The Figure 4 (a) shows the
phantom image used. To improve our results, we used the real data image [6].
The Figure4 (b) shows the real image used in this work. It is a T1-weighted
image acquired from 1.5 Tesla MRI machine (GE, Waukesha, WI) using an 8-
channel head coil 3D and spoiled gradient echo sequence (SPGR). This image
is acquired using the following parameters: TE = 8 ms, TR = 17.6 ms, an
angle of 20°, and a field of view (FOV) of 20cm × 20cm × 20cm with a size
of 200 × 200 × 200 for a isotropic resolution 1 mm3.

Fig. 4 (a) unreal data image, (b) real data image.

We applied the three compression methods: DCT, DWT and CS on images data
used. We studied the results of the three compression methods by evaluating
performance parameters such as PSNR and RLNE. We compared the three
compression methods by studying the image quality and the compression time.
We have chosen in the two compression methods: DCT and DWT different
percentages in order to evaluate different thresholds. The goal is to choose the
most correct threshold.

668
Title Suppressed Due to Excessive Length 5
Algorithm 1 explains the different steps followed in each method:

Algorithm 1:
• Load image data (phantom or MRI).
• Apply the DCT (or DWT) on the data.
• Take the absolute value of the new data.
• Select data on a table in descending order.
• Choose different percentage of the most
relevant data (1%, 5%, 10%,20%, 30% and 50%).
• Evaluate the threshold by eliminating unnecessary data.
Apply the DCT-1 (or DWT-1) on the data
obtained after the threshold.

In the compressed sensing method, we used a mask to ensure the sparsity of

the image. This mask chooses a large number of points at the centerof the
frequency data. These points correspond to the most relevant points.
Algorithm 2 explains the different steps followed.

Algorithm 2:
• Load image data (phantom or MRI)
• Apply a mask on the frequency data
• Nonlinear reconstruction to the new data.

In this part, we applied the DCT and DWT compression methods on the
phantom and real images. We used different percentages: 1%, 5%, 10%, 20%,
30%, and 50% for both DCT and DWT methods. Figures 5 and 6 represent
respectively reconstructed compression phantom and real images with DCT and
DWT methods.

Fig. 5 Images compressed by the DCT method using different percentages 1%, 5%, 10%,
20%, 30%, and 50% (a) phantom (b) real.

669
6 [Link], [Link]

Fig. 6 images compressed by the DWT method using different percentages 1%, 5%, 10%,
20%, 30%, and 50% .(a) phantom (b)real.

In this part, we applied Compressed Sensing compression method on both

phantom and real images, we used a mask showed by the Figure7 (a). Applying
the CS method on both phantom and real images, Figures 7 (b) and
(c) represent respectively phantom and real reconstructed images compressed
with the CS method.

Fig. 7 CS method using (a) mask, reconstructed compressed image (b) phantom (c) real.

After the compression of images, we noticed that the quality of the phantom
and real images reconstructed by the three methods is identical. Therefore, we
choose in following applications the real image. Moving from one method to
another, we noticed that the DCT and DWT methods give good quality images
compared to the CS method. Quantitatively, we studied the quality of the real
images by evaluating the two parameters: the PSNR and the RLNE. The Table
1 compares the different results obtained by the three compression methods
DCT, DWT and CS. The Table2 shows the time required to compressed image
for each method.

670
Title Suppressed Due to Excessive Length 7
Table 1 Evaluation parameters obtained from compressed images by three compression
methods: DWT and DCT using different percentages (1%, 5%, 10%, 30% and 50%) and CS.
PSNR RLNE
DCT 1% 27.6387 0.1837
5% 33.4647 0.0939
10% 35.2194 0.0767
20% 37.7316 0.0575
30% 40.0271 0.0441
50% 44.8865 0.0252
DWT 1% 31.0985 0.1233
5% 34.0302 0.0880
10% 35.5564 0.0738
20% 37.9702 0.0559
30% 40.2243 0.0431
50% 45.0828 0.0247
CS 30.9564 0.1254

Table 2 Compression times for the three methods

Method DCT DWT CS

Compression time (s) 9.869361 5.707984 1.004419

From results obtained by the Table 1, we noticed that the DWT method
improved compressed image quality with a high PSNR and a reduced RLNE
compared to other compression methods. Qualitatively and quantitatively, the
two images compressed by the two methods DWT and DCT give
Approximately same results. We notice that 10% of information is sufficient
to obtaina good compressed image quality witch is approximately the same of
the original image. The Figure 8 shows correctly these images. Qualitatively,
the DWT method improves the image quality much more with noise
suppression.

Fig. 8 (a) Original real data image, compressed image with 10% of data using (b) DCT
and (c) DWT, (d) compressed image using CS.

671
8 [Link], [Link]

In the CS method, the image quality degrades with noise occurring in the
background of the image. It presents a reduced PSNR and high RLNE
compared to parameters of the other methods.
We also noticed that the compression time by the CS method is reduced
(approximately 1s) compared to the other methods (approximately 5s for WT
and 9s for DCT). Reducing the compression time is very important in medical
imaging; it will serve to reduce the acquisition time. Therefore, we noticed
that the CS method is very important in medical applications

4 Conclusion

Image compression is a very large field that uses several methods. The difficulty
is to choose the most efficient method. In this paper, we noticed that the DCT
and DWT compression methods with the right choice of thresholding give
better results. We also noticed that the CS method presents a degraded image
quality but it allows a very fast compression time compared to other methods.
These results have brought us to think of improving the CS method using
different types of mask. We are also thinking of associating this method with
other methods or with methods used in this article: CS-DWT or CS-DCT.

Acknowledgements

PSNR: Peak signal-to-noise ratio

RLNE: relative error between image original and image compression.
DCT: discrete cosine transforms.
DWT: Discrete wavelet transform.
CS: compressed sensing.

References

1. Rao, K., Yip, P.: Discrete Cosine Transform, Algorithms, Advantages, Applications.
Académique Press, London (1990)
2. [Link] LIU, Lin-peiZHAIV, Ying GAO, Wen-ming LI, Jiu-fei ZHOU,” Image Compres-
sion Based on biorthogonaln wavelet Transform”, IEE Proceedings of ISCIT(2005).
3. S. Mallat, W. Liang Hwang” Singularity Detection and Processing with Wavelets” IEEE
TRANSACTIONS ON INFORMATION THEORY, VOL. 38, NO. 2, MARCH( 1992)
4. P. Telagarapu, [Link], [Link] .Prasanthi, et all.” Image Compression Us-
ing DCT and Wavelet Transformation,” International Journal of SignalProcessing, Image
Processing andPatternRecognition Vol. 4, No. 3, September,( 2011).
5. V. K. Goyal, A. K. Fletcher, and S. Rangan, “Compressive sampling and lossy compres-
sion,” IEEE Signal Processing Magazine, vol. 25, no. 2, pp. 48–96, (2008).

672
Title Suppressed Due to Excessive Length 9

6. [Link], A Bessaid.” Wavelet regularization in parallel imaging”. International Confer-

ence on Advanced Technologies for Signal and Image Processing (ATSIP), Fez, Morocco,
May 22-24, 2017. ISBN 978-1-5386-0551-6 IEEE (2017),.
7. [Link],”Compressed sensing”.IEEE [Link] [Link].52,no.4,pp.
12891306,Apr.(2006).
8. M. Lustig, D. Donoho, J.M. Pauly, ”Sparse MRI: the application of compressed sensing for
rapid MR imaging”, Magn. Reson. Med. 58 (2007) 1182–1195.(2007)
9. Lustig M, Donoho D L, Santos J M, et al. ”Compressed Sensing MRI” . IEEE Signal
Processing Magazine,vol 25,no.4,pp.72-82,(2008).
10. Nikolay Ponomarenko , Vladimir Lukin, Karen Egiazarian,[et al]”DCT Based High Quality
Image Compression”SCIA 2005, LNCS 3540, pp. 1177 – 1185, (2005).
11. Tom H. Karson, MD, Shalabh Chandra, MS, Annitta J,”JPEG Compres- sion of Dig-
ital Echocardiographic Images: Impact on Image Quality ”J AM Soc ECHOCAgDIOGR
(1995);8:306-18.
12. A. S. Lewis and G. Knowles ,”Image Compression Using the 2-D Wavelet Trans-
form”,IEEE TRANSACTIONS ON IMAGE PROCESS- ING, VOL. I. NO. 2. APRIL (1992)

673