A Comparative Study on Lossless compression mode in WebP, Better Portable Graphics (BPG), and JPEG XL Image Compression Algorithms

Full text

2021 8th International Conference on Computer and Communication Engineering (ICCCE)
978-1-7281-1065-3/21/$31.00 ©2021 IEEE
A Comparative Study on Lossless compression
mode in WebP, Better Portable Graphics (BPG),
and JPEG XL Image Compression Algorithms
Thulfiqar H. Mandeel
College of information
technology,
Imam Ja'afar Al-Sadiq
University,
Al Muthanna, Iraq
thulfiqar.hussein@sadiq.edu.iq
Muhammad Imran Ahmad
Faculty of Electronic
Engineering Tecnology,
Universiti Malaysia Perlis,
Kampus Pauh Putra,
02600 Arau, Perlis, Malaysia
m.imran@unimap.edu.my
Noor Aldeen A. Khalid
Faculty of Electronic
Engineering Tecnology,
Universiti Malaysia Perlis,
Kampus Pauh Putra,
02600 Arau, Perlis, Malaysia
aldeenabbas@studentmail.unim
ap.edu.my
Mohd. Nazrin Md Isa
Faculty of Electronic
Engineering Technology,
Universiti Malaysia Perlis,
Kampus Pauh Putra,
02600 Arau, Perlis, Malaysia
nazrin@unimap.edu.my
Abstract— On daily basis, millions of images related to the
fields of medicine, astronomy, and remote sensing are
generated. According to the sensitivity of the information, the
generated images require storing them without any pictorial
information loss. The lossless mode in image compression
algorithms didn't introduce any loss to the pictorial
information while compressing the image size by a ratio called
the compression ratio (CR). In this study, the most recent and
promising general image compression algorithms are
compared side by side in terms of CR. The tested algorithms
are WebP, BPG, and JPEG XL. The dataset used in this study
is the TESTIMAGES archive which has been used due to its
futuristic features such as high dynamic range (HDR) and high
resolution as well as the availability of both natural and
synthetic images. The results indicate that JPEG XL has the
best CR on average compared to the other two algorithms
when images with 8 bits per channel are used. Unlike the BPG
and WebP, the JPEG XL offered real compression when HDR
images, i.e., with 16 bits per channel, are feed for compression
while the other algorithms didn't support such bit depth and
generated images with 8 bits per channel.
Keywords—JPEG XL, BPG, WebP, digital image
compression.
I. INTRODUCTION
With the rise of human reliance on technology by the
day, the generated digital information is also increasing. The
reliance can take many forms, from snapping a photo of a
beautiful scene to the image collection for the training of
self-driving car systems and not stopping at the generation of
industrial and medical images. This generated information
whether being acoustic, pictorial, etc. needs to be saved in
media files before processing or transmitting them. These
media files occupy space on storage devices. With the
increase in the number of files, we are forced either to delete
less important files or purchase more storage devices.
The image in its digital format can be represented by
encoding the raw data pixel values using encoding
algorithms such as Huffman encoding, Run Length
Encoding, arithmetic encoding, etc. However, this way of
representing the images isn’t efficient in terms of file sizes
[1]. Hence to save the raw bitmap data efficiently, a
compression algorithm should be implemented.
The compression algorithms represent each pixel of the
image with less number of bits, hence, the compressed
images have lower bits per pixel (BPP). With lossy
compression, the less significant pieces of information are
discarded. The significance of these pieces of information,
generally, measured by to-what-extend they are observed by
the Human Visual System (HVS). The HVS is less sensitive
to high-frequency information and small color changes. The
chroma subsampling ( i.e. reduce color diversity in small
areas) and the high-frequency dropping are both
implemented in the commonly-used JPEG [2] image
compression algorithm. On the other hand, the lossless
compression algorithm encodes the pictorial information
with lower BPP and at the same time without losing any
information. The lossless compression on the contrary to the
lossy compression didn’t generate artifacts that accompany
the compression process. Such a feature is important in the
applications where the tempering with the original data isn’t
tolerated such as:
• Medical imaging, as any added artifacts or loss of
information of the image, can affect the assessment of
the medical information of the image [1][2].
• Satellite imaging, where the images are used to
extract the numerical indicators of the biomass, Soil
moisture, mineral mapping, etc., and any alteration to
the images give different indications [3][1][4].
• Astronomical images, where the photometric and
astrometric precision should be preserved [5]. As well
as any lossy compression artifacts may be interpreted
as the light from a star/planet or even diminish the
existing star/planet light from the images.
The medical and satellite images are not only perceived
by humans but also processed by other software tools to
extract further important information that can’t be sensed by
HVS, hence lossless compression is a must-go-to
compression mode [7]. Such applications generate a massive
amount of images daily, hence, require persistent storage
17
2021 8th International Conference on Computer and Communication Engineering (ICCCE) | 978-1-7281-1065-3/20/$31.00 ©2021 IEEE | DOI: 10.1109/ICCCE50029.2021.9467224
Authorized licensed use limited to: Universiti Malaysia Perlis. Downloaded on July 02,2021 at 13:59:17 UTC from IEEE Xplore. Restrictions apply.

upgrades. However, the research on image compression aims
to develop algorithms that achieve a higher compression rate
to reduce the need for storage devices.
The rest of the paper is organized as follows: Section 2
covers the related work. Section 3 gives a brief overview of
the tested lossless image compression algorithms. In Section
4, the information about the experiment setup, reference
software, parameters, and data set are given. Section 5
presents the results and the discussion and finally, the paper
concludes in Section 6.
II. RELATED WORKS
Indradjad et. al. [6] compared wavelet-based image
compression algorithms, namely JPEG 2000 and
Consultative Committee for Space Data Systems (CCSDS)
image compression algorithm. The test is done in the lossless
mode and by terms of CR, and compression and
decompression time. They used a dataset of five remote
sensing satellite images and the results showed roughly equal
performance by the two algorithms.
Xiao et al. [7] compared the performance of three lossless
compression algorithms which were JPEG-LS, integer
Discrete Tchebichef Transform (iDTT), and integer Discrete
Cosine Transform (iDCT) in term of the generated Bit Rates
(BR). They used miscellaneous gray level images[8], true
color images[9], and digital image dataset of cadavers in
MRI, CT, and anatomical modes [10]. While their results
didn’t show a clear winner when iDTT and JPEG-LS are
compared, they show that iDTT outperformed iDCT.
Choi et al. [11] compared baseline JPEG, JPEG 2000,
and JPEG extended range (XR) with regard to their lossy
compression. They tested the algorithms on x-ray digital
images in the matter of objective fidelity criteria such as
Mean square error (MSE) and Peak Signal to Noise Ratio
(PSNR) at different compression ratios.
Hu et. al. [12] compared three compression algorithms
which were Portable Network Graphics (PNG), Graphics
Interchange Format (GIF), and WebP with the jpeg as a
reference algorithm. Their study focused on testing these
algorithms for web applications in lossless and lossy
compression modes. Their compression was conducted on
images from Kodak Lossless True Color Image Suite, USC-
SIPI image database, and Wikipedia photos. And their result
showed that WebP outperformed PNG in the lossless
compression mode.
Jyrki Alakuijala et. al. [13] benchmarked the JPEG XL
against typical internet use cases which involve lossy
compression. Two metrics were used to evaluate the JPEG
XL, first is the perceptual visual quality, and second, the
encoding and decoding speed. The results are compared side
by side with the baseline JPEG and the intra-frame HEVC
coding system. They used ten photographic images and two
non-photographic images for algorithms evaluation.
Rusyn et. al. [4] compared lossless image encoding
techniques in remote sensing applications where the images
are feed to the algorithms as vectors each time rather than a
matrix. Their research focuses on the compression speed
when implemented on Field-Programmable Gate Array
(FPGA) systems.
Gunasheela and Prasantha [14] evaluated the
performance of various lossless satellite image compression
algorithms. These algorithms include 2D CALIC, 3D
CALIC, LUT, LAIS LUT as well as general image
compression algorithms such as JPEG-LS, Differential
JPEG-LS, JPEG 2000, Differential JPEG 2000. they used
images from Jasper ridge, Lunar lake, Cuprite, and Moffet
field. While the general algorithm like JPEG-LS showed less
complexity but didn’t utilize the special characteristics in
satellite images such as the correlation between adjacent
bands. The algorithm LAIS LUT with high complexity as a
downside represented the images with the least BPP among
all the algorithms.
III. IMAGE COMPRESSION ALGORITHMS OVERVIEW
Since the introduction of the baseline JPEG image coding
system by the Joint Photographer Experts Group in 1992,
JPEG became the common image compression tool. JPEG
offered higher compression with lower memory and
processing power requirements than the existing PNG image
format. However, the JPEG suffered from several drawbacks
at low bit-rate such as blocking artifacts [15], color banding,
ringing artifacts, as well as, no support for lossless
compression. This leads to the development of its successor
to fix these drawbacks. The development finished in the year
2000 with the approval from the international standard
organization (ISO) of the JPEG 2000 [16] standard. JPEG
2000 offered better compression than its predecessor on the
cost of added complexity. This complexity slowed down the
compression and decompression processes and by taking into
consideration the available hardware and software systems
capabilities back then, JPEG 2000 didn’t appeal as a new
image compression standard and resulted in limiting its
spread. Later, several other attempts were made such as
upgrading JPEG with new features by extending the range of
the color channels in JPEG XR [17] to HDR and adding
alpha channel and HDR with the same compression rate in
JPEG XT [18]. Both algorithms didn’t succeed in replacing
the baseline JPEG because the added features aren’t
convincing the vendors and end-users enough to switch to
these proposed algorithms, especially when the codex
requires implementation on hardware for faster
encoding/decoding. A new JPEG XL [13][19][20][21] was
designed in a way to avoid the drawbacks that hindered the
JPEG 2000, JPEG XT, and JPEG XR from replacing
baseline JPEG as a popular image compression standard. The
detail for JPEG XL can be found in [20] but it can be briefly
summarized as
• Color transform phase: Palette, YCoCg,
SubtractGreen, and YCbCr
• Reversible Non-linear Haar (squeeze)
• Adaptive quantization (with stored remainder)
• Prediction: left, median, average, and top
• Meta-Adaptive context model
• Entropy coding: BEGABRAC, Brotli, and ANS
JPEG XT and JPEG XR were introduced to cover the
needs for compressing HD pictures while the HEVC was
introduced to provide high performance and efficient video
compression algorithms for videos with high resolutions,
e.g., 2k, 4k, or 8k resolution. The video compression
18
Authorized licensed use limited to: Universiti Malaysia Perlis. Downloaded on July 02,2021 at 13:59:17 UTC from IEEE Xplore. Restrictions apply.

algorithms provide compression based on inter- and intra-
frame redundancy. The first frame (i.e., picture) in the video
is compressed using spatial prediction from region-to-region
in the same frame. As result, the part that can perform intra-
frame compression is used for image compression
[22][23][24].
The BPG is a wrapper for the intra-frame compression
part of the HEVC [25][26][27]. The HEVC is under many
patents which limit the BPG spread as the developers and
System on chip (SoC) vendors have to pay for the
implementation of patented algorithms.
The WebP image compression algorithm work in two
modes the lossy and the lossless compression mode. The
lossy compression is based on the intra-frame block
prediction compression that exists in the VP8 video
compression algorithm. The lossless compression mode
compresses the images by using, if applicable, the spatial
prediction, a red and blue prediction from green color, color
indexing (when the image contains less than 256 unique
colors), and lastly color cache coding for indexing colors
from the regularly updated palette. This acquired information
is then encoded with an entropy encoder for extra
compression [28] [29].
The WebP suffers from the drawback that it offers
limited resolution support of which makes
it obsolete in the future as the need for higher resolution is
already in demand and expected to keep rising. Moreover,
The BPG and the WebP are both based on video codec intra-
frame compression which makes them unaccustomed to
lossless image compression by design as the user didn’t
notice the fine details in video frames.
IV. TESTING SETUP
A. Test environment
The experiments are performed on the same system set
up to ensure the test results are justly compared in the same
condition. The operating system used for the testing was 64-
bit windows 10 Home installed on a laptop computer with
Intel Core i7-6700HQ CPU @ 2.60GHz and 16.0 GB of
RAM. At the time of conducting this research, JPEG XL
executables run only through virtualization using Docker
software with Ubuntu Linux image which rolled out the
speed test as the platforms differ and there won’t be a fair
comparison.
B. Reference software and commands
The reference software programs that used for testing are:
• jpeg xl source code for a JPEG XL image format that
includes updates up to 12/October/2020 was
downloaded from the official channel [30] and was
compiled and executed under Docker container
according to the official guide. The encoding command
is:
./cjpegxl -q 100./inputIMG.png./outputIMG.jxl
the parameter q is quality setting for a modular mode in
which its value should be in the range: 0-100 where 100
means mathematically lossless
• bpg-0.9.8-win64 executables for BPG format obtained
from the developer website [25]. The encoding
command is:
.\bpgenc.exe -lossless '.\inputIMG.png' -
o.\outputIMG.bpg
the parameters: lossless to enable lossless mode, o to set
output filename. the default parameters are ycbcr for color
space, 8 bits per channel (another available bit depth is 12),
and x265 as HEVC encoder.
• libwebp-1.1.0-windows-x64 executables for WebP
image format obtained from the developer website [31].
The encoding command is:
.\cwebp.exe -lossless.\inputIMG.png -o
outputIMG.webp
the parameters: lossless is to encode the image
losslessly, o to set output filename
C. DATASET
As the algorithms tested for their lossless mode which
means no dependence on the characteristics of the pictorial
information, a non-domain-specific images dataset can be
used. There are a large number of datasets on the internet
that can be used to test the image processing algorithms such
as UCID [33]. However, these datasets suffer from two
major drawbacks that affect their usability in digital image
processing research. The first drawback is the licensing and
copyright limitation. The second drawback is that the free
datasets are homologous in terms of color band, resolution,
gradients, and pattern and usually not up to date in
technological terms [34]. To overcome these drawbacks,
Asuni & Giachetti [34] proposed a TESTIMAGES archive
that resolves the aforementioned issues so it can become a
reference dataset for testing image processing algorithms.
Their datasets provide several features that are missing from
the existing datasets which are:
• Each image available in the original format besides
a downsampled and/or shifted versions with 8 and 16 bits per
channel in 3 Channels Red-Green-Blue (RGB) and 1
Channel grayscale format.
• The support for modern technological criteria such
as high resolution and HDR.
• All the images are copyrighted under Attribution-
NonCommercial-ShareAlike 4.0 International (CC BY-NC-
SA 4.0).
• The images are saved using PNG format which is
lossless, paten-free, and accepted as input for various image
compression software programs.
The dataset contains different sub-categories that differ in
a characteristic. These sub-categories are:
1.1.1 SAMPLING
This category is representing photographic images. Each
base image is 2400x2400 pixels, 16 bpp, HDR, with RGB
color space. The other images are obtained from the base
image by reducing the bpp to 8, changing the color space to
grayscale, and reducing the resolution by factors 2, 3, 4, 5, 6,
8, 10, 12, 15, 16, 20 and 24. The images have a maximum
HDR intensity level equal to . An
example of the images from this category is shown in Fig. 1.
19
Authorized licensed use limited to: Universiti Malaysia Perlis. Downloaded on July 02,2021 at 13:59:17 UTC from IEEE Xplore. Restrictions apply.

Fig. 1: The sample from the dataset used for testing which are: (a)
billiard_balls_a, (b) chairs, (c) carrots, (d) lion, (e) tomatoes_a, (f) scarf
1.1.1 PATTERNS
This category of images contains computer-generated
simple geometrical patterns. The reference images are
grayscale, with 16 bits per channel, and color images version
can be generated by copying the grayscale values into one or
more of the color channels. This dataset is used to check the
compression algorithm performance when non-textured
images with a limited set of colors are used as an input. Fig.
2 shows samples from images in this category.
Fig. 2. The sample from the dataset used for testing which are: (a)
bars_45deg_bltr_0064, (b) bars_horizontal_sin_1024_3xhalfpi, (c)
circles_center_0032, (d) grid_circles_progressive_a, (e) grid_triangle_a, (f)
grid_triangle_b
D. Testing criteria
In the contrast to the subjective test, which involves
human subjective assessment for the quality of the generated
image, the objective test consists only of numerical
measurements. As the scope of this research is solely lossless
image compression mode which didn’t alter the pictorial
information, hence only the CR test is conducted. The CR
determines how well an algorithm has compressed an image
through dividing the size of the original image by the size of
the compressed image [32]. Equation (1) demonstrates this
concept:
CR = U/ C (1)
Where CR is the compression ratio, U is the size of the
image before compression, while C is the file size after
compression. Both sizes of the image before and after
compression are measured in Bytes.
V. RESULTS AND DISCUSSION
In this Section, the CR results from testing the image
compression algorithms are presented. First, the Shannon
entropy is presented to show the entropy of the images used
in the test in Fig. 3 and Fig. 6 for the categories SAMPLING
and PATTERN respectively. The Shannon entropy is used to
measure the amount of the information in PATTERN and
SAMPLING images. Hence, measure the image compression
performance of the algorithms when less- and high-
complexity images are present.
When the SAMPLING category dataset is used as input,
the CR results showed in Fig. 4 and Fig. 5 while the CR
results for the PATTERN category dataset are shown in Fig.
7 and Fig. 8. The purpose of using two resolutions for the
test is to measure the effects of the resolution on the
performance of the algorithms.
When comparing the results from Figures 4 and 5 and
Figures 7 and 8, we can see that all the algorithms showed
relatively better CR in images that have lower entropy, i.e.
PATTERN images, compared to the images with higher
entropy, i.e. SAMPLING. This can be traced to the diverse
pictorial information in the SAMPLING images while the
PATTERN images contain only simple geometrical patterns.
The WebP, as well as BPG, are both support generating
images with 8 bits per channel while JPEG XL produces
images with 16 bits per channel. Hence, in this case, BPG
and WebP will produce misleading higher CR while JPEG
XL produces true lossless compression. For a fair
comparison, testing on images with 16 bits per channel
didn’t get carried out.
WebP embodies various transformations that can be
selected beforehand depending on which transform will yield
a minimum entropy. This feature enables WebP to deal with
miscellaneous types of images efficiently. Each transform in
the WebP produces a different CR. One of these transforms
namely the color indexing transform provides a superior
result when dealing with synthetic images that generally
having several unique alpha-red-green-blue (ARGB) values
which can be seen in Figures 7 and 8. There's also another
feature that increases the CR in WebP which is the local-
orientation-based prediction selection. As different parts of
the image have different gradients this adaptive prediction
model is more efficient than a fixed prediction model.
As in WebP, the JPEG XL also provide an adaptive
prediction model depends on the bi-directional nature of the
block that is currently being processed. during the entropy
encoding, the JPEG XL has the advantage of using an
Asymmetric Numeral System (ANS) which unlike the
Huffman coding, has the same CR as Arithmetic Coding but
with superior speed improvement. The entropy encoding is
furthermore supported by the context model which is used to
separate entropy sources and hence increase the entropy
encoding efficiency. The JPEG XL produced a higher CR
compared to the other algorithms when high-resolution
images are used.
The BPG was designed for video compression which
requires minimum computation per frame. Moreover, the
video compression didn't require keeping many details as the
20
Authorized licensed use limited to: Universiti Malaysia Perlis. Downloaded on July 02,2021 at 13:59:17 UTC from IEEE Xplore. Restrictions apply.

end-user didn't pause the videos so often to observe fine
details. Hence, it is tailored for lossy compression than
lossless compression. The results in Figures 4, 5, 7 and 8
imply this notion as it exhibited the least CR among the
tested image compression algorithms.
The JPEG XL includes different algorithms in each
compression phase which make it adaptive to the varios kind
of input images; hence, yield best performance in different
cases.
Fig. 3: Shannon entropy for SAMPLING images with 480x480 dimensions
and 3x8bit
Fig. 4: CR for SAMPLING images with 480x480 dimensions and 3x8bit
Fig. 5: CR for SAMPLING images with 2400x2400 dimensions and 3x8bit
VI. CONCLUSION
This research tested the lossless compression mode in
three image compression algorithms which were JPEG XL,
BPG, and WebP. TESTIMAGES dataset which has modern
features i.e., high image resolution and 16 bits per channel
was used. This research demonstrates that the WebP has a
slightly better CR than JPEG XL on synthetic images with
simple geometrical patterns, limited ARGB unique values,
and low resolution. However, the JPEG XL produces the
best results on the other three tests when natural images are
used as well as high-resolution synthetic images. The BPG
showed poor CR performance in all four tests. The CR
results indicate the suitability to invest in the newly
introduced image compression algorithm i.e., JPEG XL as a
cutting-edge image compression algorithm for the
applications of medicine, astronomy, and remote sensing.
Shannon Entropy
0 2 4 6 8
bars_45deg_bltr_0064.png
bars_horizontal_sin_0256_0xhalfPI.png
circles_center_0032.png
grid_circles_progressive_a.png
grid_triangle_a_0064x0048_0.png
grid_triangle_b_0080x0060_1.png
Fig. 6: Shannon entropy for PATTERNS images with 640x480 dimension
and 3x8bit
Fig. 7: CR for PATTERNS images with 640x480 dimension and 3x8bit
Fig. 8: CR for PATTERNS images with 2880x1620 dimensions and 3x8bit
21
Authorized licensed use limited to: Universiti Malaysia Perlis. Downloaded on July 02,2021 at 13:59:17 UTC from IEEE Xplore. Restrictions apply.

REFERENCES
[1] K. Sayood, Introduction to Data Compression, vol. 5, no. 2. .
[2] D. Yee, S. Soltaninejad, D. Hazarika, G. Mbuyi, R. Barnwal, and A.
Basu, “Medical image compression based on region of interest using
better portable graphics (BPG),” 2017 IEEE Int. Conf. Syst. Man,
Cybern. SMC 2017, vol. 2017-Janua, pp. 216–221, 2017, doi:
10.1109/SMC.2017.8122605.
[3] R. C. Gonzalez and R. E. Woods, Digital Image Processing (4th ed.).
2018.
[4] B. Rusyn, O. Lutsyk, Y. Lysak, A. Lukenyuk, and L. Pohreliuk,
“Lossless image compression in the remote sensing applications,”
Proc. 2016 IEEE 1st Int. Conf. Data Stream Min. Process. DSMP
2016, no. August, pp. 195–198, 2016, doi:
10.1109/DSMP.2016.7583539.
[5] W. D. Pence, R. Seaman, and R. L. White, “Lossless Astronomical
Image Compression and the Effects of Noise,” Publ. Astron. Soc.
Pacific, vol. 121, no. 878, pp. 414–427, 2009, doi: 10.1086/599023.
[6] A. Indradjad, A. S. Nasution, H. Gunawan, and A. Widipaminto, “A
comparison of Satellite Image Compression methods in the Wavelet
Domain,” IOP Conf. Ser. Earth Environ. Sci., vol. 280, no. 1, 2019,
doi: 10.1088/1755-1315/280/1/012031.
[7] B. Xiao, G. Lu, Y. Zhang, W. Li, and G. Wang, “Lossless image
compression based on integer Discrete Tchebichef Transform,”
Neurocomputing, vol. 214, pp. 587–593, 2016, doi:
10.1016/j.neucom.2016.06.050.
[8] “CVG - UGR - Image database - Miscelaneous gray level images
(256 x 256).” http://decsai.ugr.es/cvg/dbimagenes/g256.php (accessed
Nov. 09, 2020).
[9] “Kodak Lossless True Color Image Suite.”
http://r0k.us/graphics/kodak/ (accessed Nov. 09, 2020).
[10] “The Visible Human male and female datasets.”
https://www.nlm.nih.gov/research/visible/getting_data.html (accessed
Nov. 09, 2020).
[11] H. R. Choi, S. H. Kang, S. Lee, D. K. Han, and Y. Lee, “Comparison
of image performance for three compression methods based on digital
X-ray system: Phantom study,” Optik (Stuttg)., vol. 157, pp. 197–202,
2018, doi: 10.1016/j.ijleo.2017.11.069.
[12] J. Hu, S. Song, and Y. Gong, “Comparative performance analysis of
web image compression,” Proc. - 2017 10th Int. Congr. Image Signal
Process. Biomed. Eng. Informatics, CISP-BMEI 2017, vol. 2018-
Janua, pp. 1–5, 2018, doi: 10.1109/CISP-BMEI.2017.8301939.
[13] J. Alakuijala et al., “Benchmarking JPEG XL image compression,” p.
32, 2020, doi: 10.1117/12.2556264.
[14] K. S. Gunasheela and H. S. Prasantha, “Satellite image compression-
detailed survey of the algorithms,” Lect. Notes Networks Syst., vol.
14, pp. 187–198, 2018, doi: 10.1007/978-981-10-5146-3_18.
[15] Y. L. Lee, H. C. Kim, and H. W. Park, “Blocking effect reduction of
JPEG images by signal adaptive filtering,” IEEE Trans. Image
Process., vol. 7, no. 2, pp. 229–234, 1998, doi: 10.1109/83.661000.
[16] D. Santa-Cruz and T. Ebrahimi, “An analytical study of JPEG 2000
functionalities,” IEEE Int. Conf. Image Process., vol. 2, pp. 49–52,
2000, doi: 10.1109/icip.2000.899222.
[17] F. Dufaux, G. J. Sullivan, and T. Ebrahimi, “The JPEG XR image
coding standard [Standards in a Nutshell],” IEEE Signal Process.
Mag., vol. 26, no. 6, pp. 195–200, 2009, doi:
10.1109/MSP.2009.934187.
[18] P. Hanhart and T. Ebrahimi, “Evaluation of JPEG XT for high
dynamic range cameras,” Signal Process. Image Commun., vol. 50,
no. October 2016, pp. 9–20, 2017, doi: 10.1016/j.image.2016.10.004.
[19] A. Rhatushnyak et al., “JPEG XL Image Coding System,” ISO/IEC
JTC 1/SC 29/WG 1. 2019.
[20] J. Sneyers, J. Wassenberg, and J. W. Jyrki Alakuijala, Joao Ascenso,
Sami Boukortt, Martin Bruse, Iulia-Maria Comsa, Touradj Ebrahimi,
Moritz Firsching, Thomas Fischbacher, Sebastian Gomez, Walt
Husak, Renata Khasanova, Evgenii Kliuchnikov, Robert Obryk,
Fernando Pereira, Antonio Pinheiro, K, “JPEG White Paper : JPEG
XL image coding system.” jpeg.org.
[21] J. Alakuijalaa et al., “JPEG XL next-generation image compression
architecture and coding tools,” 2020, no. September 2019, doi:
10.1117/12.2529237.
[22] G. J. Sullivan, J. R. Ohm, W. J. Han, and T. Wiegand, “Overview of
the high efficiency video coding (HEVC) standard,” IEEE Trans.
Circuits Syst. Video Technol., vol. 22, no. 12, pp. 1649–1668, 2012,
doi: 10.1109/TCSVT.2012.2221191.
[23] U. Albalawi, S. P. Mohanty, and E. Kougianos, “A Hardware
Architecture for Better Portable Graphics (BPG) Compression
Encoder,” Proc. - 2015 IEEE Int. Symp. Nanoelectron. Inf. Syst. iNIS
2015, pp. 291–296, 2016, doi: 10.1109/iNIS.2015.12.
[24] J. Lainema, F. Bossen, W. Han, J. Min, and K. Ugur, “Intra Coding of
the HEVC Standard,” IEEE Trans. CIRCUITS Syst. VIDEO
Technol., vol. 22, no. 12, pp. 1792–1801, 2012.
[25] Fabrice Bellard, “BPG Image format.” https://bellard.org/bpg/
(accessed Oct. 08, 2020).
[26] M. P. Sharabayko, O. G. Ponomarev, and R. I. Chernyak, “Intra
compression efficiency in VP9 and HEVC,” Appl. Math. Sci., vol. 7,
no. 137–140, pp. 6803–6824, 2013, doi: 10.12988/ams.2013.311644.
[27] J. Lainema, M. M. Hannuksela, V. K. M. Vadakital, and E. B. Aksu,
“HEVC still image coding and high efficiency image file format,”
Proc. - Int. Conf. Image Process. ICIP, vol. 2016-Augus, pp. 71–75,
2016, doi: 10.1109/ICIP.2016.7532321.
[28] “Compression Techniques | WebP | Google Developers.”
https://developers.google.com/speed/webp/docs/compression
(accessed Oct. 10, 2020).
[29] “WebP Lossless Bitstream Specification | Google Developers.”
https://developers.google.com/speed/webp/docs/webp_lossless_bitstre
am_specification#4_image_data (accessed Nov. 04, 2020).
[30] “jpeg / JPEG XL Reference Software · GitLab.”
https://gitlab.com/wg1/jpeg-xl (accessed Nov. 09, 2020).
[31] “A new image format for the Web | WebP | Google Developers.”
https://developers.google.com/speed/webp (accessed Nov. 09, 2020).
[32] I. C. Techniques, A. Citation, and I. C. Techniques, Image
Compression Techniques : A Survey in Lossless and. 2018.
22
Authorized licensed use limited to: Universiti Malaysia Perlis. Downloaded on July 02,2021 at 13:59:17 UTC from IEEE Xplore. Restrictions apply.