Springer Nature 2021 LATEX template

Combined Run-Length and Huffman

Encoding for Image Compression
Yaroslav Krainyk1*
1* Computer Engineering Department, Petro Mohyla Black Sea
National University, 10. 68 Desantnykiv, Mykolaiv, 54003,
Ukraine.

Corresponding author(s). E-mail(s):

yaroslav.krainyk@chmnu.edu.ua;

Abstract
The paper introduces combined Run-Length encoding (RLE) and Huff-
man encoding approach for image compression. The approach can be
applied to the peculiar type of images that satisfies specific characteris-
tics in terms of color components data. The devised approach is supposed
to implement lossless compression for the preprocessed image data. First,
the RLE is applied to the prepared data, then, Huffman coding process
is performed. The results received during the test session of the imple-
mented software showcase that the proposed technique is feasible and
provides a decent level of compression. It can be implemented also in the
hardware since the complexity of the operations used is relatively low.

Keywords: compression, Huffman encoding, Run-Length encoding

1 Introduction
The ubiquitous presence of multimedia data has greatly extended the user
experience in modern devices Nauman et al (2020). One of the main catalysts
of this trend is the continuous improvement of the compression algorithms
as demonstrated in Lee and Kim (2018) and ZainEldin et al (2015). This
process elevates the performance by reducing the size of the media data to be
transmitted via communication channel. The decompression is performed on
the receiving device that, eventually, is responsible for the output (image on the

1
 Springer Nature 2021 LATEX template

2 Combined Run-Length and Huffman Encoding for Image Compression

screen, sound in the headphones, etc.). Hence, the development of multimedia

technologies has a prerequisite in the form of compression techniques that are
involved in the technology.
On the other hand, data structure to be stored often dictates the selection
of the appropriate method and incites development of the new one Sayood
(2017). While some algorithms may be regarded as universal in terms of what
data type they can be applied to, it is impossible to achieve the optimal com-
pression efficiency on different data sources using only one algorithm. The
primary reason for this effect is that information storage formats are different
by their structure. Hence, it is reasonable to rely on the knowledge of the data
structure to supply better compression solutions. In this work, the specific type
of preprocessed image is considered as an object of interest. The internal file
organization is exploited to provide compression of the image.
In this work, combined Run-Lenght Encoding-Huffman (RLE-Huffman)
technique for image compression is proposed that introduces the following
novelties:
• RLE is used to compress a single specific value that is the most common
value that appears in the image color components;
• To avoid possible overgrowth of the encoded sequences, only sequences that
contains number of elements that is higher than their encoded substitution
will be encoded;
• Additionally, the length of the sequence is limited by minimal and maximum
value.
• The rest values of the color components and compressed sequence block
undergo compression using Huffman algorithm; two options are possible
to implement this 1. Usage of precomputed substitution table 2. Dynamic
computation of Huffman tree and creation of the specific substitution table.
• Devised the structure of the compressed file that makes possible further
decompression.
The rest of the paper is organized as follows. The Related work review
section contains the review of the current compression approaches that are
utilized in the multimedia realm. It also establishes the background of mul-
timedia data compression in general. The Combined compression approach
introduces the proposed compression technique that is based on the combina-
tion of Run-Length Encoding (RLE) and Huffman encoding. The Results and
Discussions section overviews the results that were attained by using the com-
bined approach. The Conclusion section generalizes the findings in context of
the proposed solution and proposes possible outsights for further development.

2 Related work review

Without a shadow of a doubt the topic of image and multimedia compres-
sion can be named as a top notch topic of interest for scientific research.
 Springer Nature 2021 LATEX template

Combined Run-Length and Huffman Encoding for Image Compression 3

Immense amount of publications has indicated significant focus on the realm of

multimedia compression, e.g. Kwan et al (2019), Elakkiya and Thivya (2021).
The JPEG algorithm Wallace (1992) is a cornerstone for image compres-
sion. Not only it established a foundation for highly efficient lossy compression
format, but also served as a basis for further developments in the area. The
processing of the image data passes through several stages during compression.
First, the colorspace conversion is performed and YCrCb colorspace is used
for storage. The most important step that allows achieving high coefficient of
compression is application of the Discrete-Cosine Transform (DCT) Ahmed
et al (1974) to the parts of the image with size of 8x8 pixels. This makes possi-
ble further compression performed by run-length coding and Huffman coding
Knuth (1985). As can be observed, from even a brief description of JPEG
format, it incorporated multiple techniques that are applied sequentially to
provide better compression results. The idea of the presented approach is also
based on leveraging encoding by applying sequentially two algorithms.
The appearance of JPEG format greatly facilitated developments in the
field of video compression. Particularly, the widespread MPEG format Schwarz
et al (2018) exploits techniques introduced in the image format and enhances
them with regard to the video format details. The default image storing format
in many user devices is JPEG. The main reason for this is that it provides a
reasonable compromise between compression ratio and image quality.
Another option that is abundantly used in image compression is the fam-
ily of Lempel-Ziv (LZ) algorithms Apostolico (2007). They are based on the
concept of a “sliding” window where the next symbol (or sequence of symbols)
can be encoded using information that appeared in the file previously. The
same principle applies also for decoding procedures. LZ algorithms belong to
the category of lossless compressors. The LZ compression algorithm does not
work on some peculiar block of the image but rather sequentially processes
input data over the whole image canvas (row by row). Therefore, two general
options for compression organization are possible:
1. Block processing.
2. Sequential processing.
Combination of the methods can be also used. In this work, the sequential
processing was applied.
One of the most advanced contemporary compression libraries, zstandard
or zstd Facebook (2022), was implemented with the idea of its universal appli-
cation. In fact, this is a highly configurable library that ensures possible
compromises between compression size and compression speed. The decision
about this can be made according to the user preferences. The core of the
library actually implements different algorithms with a combination of LZ77
and Finite State Entropy and Huffman encoding Wang (2019) as the preva-
lent coding approach. The algorithm that is to be applied depends on the
supplied parameters. For instance, parameter value 22 ensures the best com-
pression ratio but with most resources consumed. Due to the decompression
 Springer Nature 2021 LATEX template

4 Combined Run-Length and Huffman Encoding for Image Compression

speed, zstd found various applications including being a part of Linux kernel
for btrfs. Another point to mention about zstandard is that it is lossless.
While different compression algorithms demonstrate their presence in dif-
ferent data types compression applications, the general recommendation for
the encoding process is to exploit peculiarities of data structure to achieve
better results. For instance, data with repetitive chunks located on a small
offset from each other is extremely suitable for LZ coding. RLE is an obvious
candidate to encode long sequences of a single repetitive value. On the other
side, if the length of sequences is not too big, it can cause an increase of the
encoded file. Therefore, it is necessary to take into account the structure of the
incoming data for the compression procedure. As can be observed even from
the brief description of JPEG and zstd, they are not constrained with only
one compression algorithm and employ different techniques to achieve better
results.
The combination of RLE encoding and Huffman compression is relatively
common. In Stabno and Wrembel (2009), authors applied the technique to the
compression of database information. Additionally, as can be observed, the
use of Huffman encoder at the final stage is quite common. The explanatory
factor for this decision is that it has relatively low complexity and adds up to
the compression level achieved at the previous stage. However, it is not always
guaranteed so this additional stage should be used cautiously. The proposed
approach is oriented on image compression and declares the difference between
the previous findings in the structure of the data and details of the compression
mechanism.

3 Combined compression approach

The proposed compression approach is supposed to be applied on the part of
the preprocessed image. The preprocessing procedure actually separates image
data into two parts:
1. Residual values.
2. Node values.
Node values carry information about the initial image. Node values are
taken from the initial image with the predefined step, for instance, each eighth
pixel value in both dimensions. The residual values are the values that actually
undergo the procedure of removing the least significant bits and centering
around the middle value. Due to this fact, the histogram of the residual values
can be expected to follow the same histogram pattern. Encoding and decoding
of the residual values is the main point of interest of the paper. Further in
the paper they are considered as an individual image separated from the node
values.
Typically, the information about the image is stored using three color
components, e.g., Red-Green-Blue (RGB). In the uncompressed presentation,
values for color components alternate with a fixed small step. Because the
 Springer Nature 2021 LATEX template

Combined Run-Length and Huffman Encoding for Image Compression 5

correlation between the respective color components in the neighboring pix-

els is higher than between individual color components inside a pixel, it is
convenient to consider sequences of color components and apply compression
to them. For instance, the red component sequence is encoded independently
from green and blue components. This data arrangement has been also applied
in the presented work.
The paper deals with the compression of images that have one prevalent
value of the color component for all color components. Moreover, the distribu-
tion of the color components values should comply with histogram with one
expected value and the frequency of other values decreasing as they become
distant from the center value. These two requirements will allow the coder to
apply RLE and Huffman techniques effectively.
The compression approach proposed in this paper incorporates use of two
algorithms for compression:
1. Run-Length Encoding - to compress long sequences of the repetitive data
values.
2. Huffman encoding - to exploit the feature of the color component histogram
of the values in the image after RLE is applied.
At the first step, all sequences of the repetitive selected value are supposed
to be compressed using RLE. They are stored in the following way. To represent
the sequence, four bytes are allocated. The first two bytes represent the offset
and the second pair of bytes encodes the length of the sequence. For this case,
the minimal length of the sequence to guarantee some improvement in the
output file size is selected to be more than 4. Otherwise, it may result in the
opposite effect, increased file size if the image contains many short sequences.
There is also an implication about the maximum length of sequence that can
be encoded this way and it is limited to the value of 65536. Hence, if the actual
sequence has a length that is higher than this value, it may be encoded as
the sequence of the encoded RLE blocks. Obviously, this is the most desired
scenario for this type of image compression.
The Huffman encoder belongs to the category of entropy encoders. It relies
on the fact that some symbols appear in the file more often than others.
Those symbols can be encoded with shorter bit sequences while symbols that
occur rarely may be assigned to a longer bit sequence. The Huffman algorithm
supposes building of a binary tree for encoding or the tree can be precom-
puted in advance if the histogram of values is known in advance. Therefore, it
can be stored as a look-up table. Notice that all RLE-encoded sequences also
undergo the procedure of Huffman coding. That fact entails changes in the
values distribution because:
1. RLE sequence is a pair of two-byte values that are treated as 16-bit words
at the previous stage.
2. At current point they should be regarded as a set of four byte values. In the
opposite case, the structure of the Hamming encoder increases significantly
due to the increased number of table entries.
 Springer Nature 2021 LATEX template

6 Combined Run-Length and Huffman Encoding for Image Compression

It cannot be assumed that some values will not appear. Thus, the table
should contain entries for all of the values in the range of 0. . . 255. The exten-
sion of the dictionary to store not only bytes but words is not feasible in this
case because it would significantly increase complexity of the system.
There two options available for Hamming compression:
1. Compute the statistics in real-time, construct Huffman binary tree and then
construct a look-up table.
2. Use of a precomputed table for all images.
In this work, the second option is considered as preferable because it is
assumed that the distribution of the values remains close to the initial one.
For this purpose, investigation on the color components values histogram was
executed to provide necessary background to the assumption. According to
the extracted data, the histograms follow the similar pattern. The examples
are shown in the Figures 1, 2, 3, 4 below.

100000

80000

60000
Number of occurences

40000

20000

0
0 50 100 150 200 250
Color component value

Fig. 1 Color values distribution

As can be seen from the histograms, the histogram of values has one central
value (128) which is supposed to be compressed using RLE. In the remainder
of the paper, the value 128 is selected as the one that RLE is applied on. The
histogram of other values can be well represented by a binomial distribution.
However, some values are missing from one distribution to another. Therefore,
the following technique can be used if some values do not appear. The number
of their appearances is assigned to the minimal number of values that is not
equal to 0, but this means that the word length for the symbol in Hamming
tree may be assigned randomly and result in poor performance. Another obser-
vation about the distribution is that the width of the central part is different
for different images. Additionally, in general, it descends in both directions
from 128. Thus, gradually decreasing values can be assigned to the number of
 Springer Nature 2021 LATEX template

Combined Run-Length and Huffman Encoding for Image Compression 7

4500000

4000000

3500000

3000000
Number of occurences

2500000

2000000

1500000

1000000

500000

0
0 50 100 150 200 250
Color component value

Fig. 2 Color values distribution

2500000

2000000

1500000
Number of occurences

1000000

500000

0
0 50 100 150 200 250
Color component value

Fig. 3 Color values distribution

occurrences with 0 and 255 having the smallest number of occurrences. Dur-
ing the construction of the Hamming table these values should be assigned to
the longest words in the table.
Another observation received from the histograms is that the content of the
files is extremely suitable for Huffman compression. And while the common
value is encoded mostly using RLE, the rest of the file, that also includes
shorted sequences of the common value, should fit in the context of Huffman
algorithm application.

3.1 Structure of the compressed file

In order to avoid any ambiguity during the decoding process, the structure
of the compressed file has to be known in advance. Hence, to ensure this
 Springer Nature 2021 LATEX template

8 Combined Run-Length and Huffman Encoding for Image Compression

7000000

6000000

5000000
Number of occurences

4000000

3000000

2000000

1000000

0
0 50 100 150 200 250
Color component value

Fig. 4 Color values distribution

requirement, it was proposed to always have the RLE-encoded sequence at the

beginning of the file. Furthermore, the same is applicable to the file ending.
The maximum overhead of this operation is use of only 8 excessive bytes.
This results in the completely deterministic structure of the compressed file.
However, it is never guaranteed that the file actually starts and ends with a
sequence of the same values. Thus, the idea of this configuration is to store
information about where the next compressed sequence is located. Even if the
file contains no sequence at the beginning, it can be encoded as a sequence with
zero length and a maximum offset. The visual representation of the described
file structure is shown in the Fig. 5.
As can be seen from the Figure, the file structure is organized as a set of
linked RLE-blocks with residual values that alternate those blocks. Sometimes,
residual parts can have zero length. In this case, two RLE-blocks occur imme-
diately after one another. Apparently, the file that consists only of RLE-blocks
can be encoded this way, as well as the file that contains RLE-blocks with
zero-only length (no sequence can be encoded; only residual blocks contain
initial information about image). As for the multiple color components that
can represent a pixel, the sequences can be stored one after the other. Pres-
ence of multiple color components enforces the developer to add information
about the sequence length. Otherwise, the borders of two different sequences
may become undistinguishable. Hence, the easiest solution is to add the length
at the beginning of the file. The assumption is that four bytes at the start of
the file can be dedicated to store the length of the sequences or the number of
pixels in the uncompressed image.
As for the Hamming compression, the Hamming table is necessary to
decode the whole file. Therefore, it is also stored in the file to provide an all-
around application. However, if there is a convention that only one table is
used across all files, then the table can be omitted.
 Springer Nature 2021 LATEX template

Combined Run-Length and Huffman Encoding for Image Compression 9

Fig. 5 Intermediate structure of the file after RL encoding stage

Additionally, it can be allegedly stated that the proposed file structure

facilitates streaming capability of the code. Providing that the encoder works
on the stream of data without preliminary knowledge of the sequence loca-
tions, a buffer for incoming data is necessary. As long as the encoder registers
the prevalent value in the input with necessary length, it generates RLE-block
as an output. However, as two types of blocks are always alternating, hard-
ware implementation of the encoder should actually comprise two buffers for
both sequence types. The depth of the buffer is chosen in advance. In the
implemented version of the method, the buffer is limited to the 65536 elements
because this is the maximum number of elements that can be addressed by
two bytes. The outputs of the buffers are connected with Huffman encoder
block that just processes incoming data. The control over the multiplexer that
selects incoming data is supposed to be realized in the top level logic of the
hardware stream encoder.
 Springer Nature 2021 LATEX template

10 Combined Run-Length and Huffman Encoding for Image Compression

3.2 Decompression procedure

The decompression assumes application of the Huffman table first and, then,
decoding of the sequences and final alignment of data in the deflated file. While
the Huffman decoding should be considered as trivial, the decoding of RLE
blocks is the key to correct result output. During this stage, two different types
of sequences are recognised:
1. Decoded RLE sequence.
2. Sequence of residues.
During the decompression procedure, the reason for the specific file align-
ment becomes apparent. The fact that the file always starts with an RLE-block
assures that the first four symbols are occupied by it. After decoding these
values, the location of the next block is known and residues can be decoded
solely by Huffman algorithm. As can be observed, RLE-blocks and residue
blocks alternate during decompression. Those actions can be implemented in
the loop with special treatment of specific cases (zero-length RLE block or last
blocks for both residues and RLE).
The decompression procedure is quite low-demanding in terms of resource
usage (processor and memory required). Moreover, it is suitable for streaming
implementation. Due to the proposed internal organization of the data, the
decoding can be started as soon as a new chunk of data arrives. Each incom-
ing symbol is decoded by Huffman algorithm and, then, can be immediately
converted to the output symbol if it is a residue symbol or generate a com-
mand to set continuous output of sequence if it is a part of RLE-block. If the
decompressor encounters RLE-block, it holds the rest of incoming data in the
designated buffer or suppresses input stream by latching the corresponding
output signal.

4 Results and Discussions

In order to verify the proposed concept, the software implementation of the
approach has been developed. It was implemented using C++ language. The
source code of the project is available at the repository Krainyk (2022). It is
a console application that accepts file name as an input parameter. The other
supported parameter is a filename of the symbol frequency. If it is omitted,
the frequency table will be built during the runtime of the program according
to the content of the file. The example of the values distribution file is shown
in the Figure 6.
The software performs decompressing immediately after the encoding is
done for the convenience of the user. The decompression operation has also
been implemented to check reversibility of the performed operations. The test
strategy was based on the extraction of the raw data from an image file with the
following compression procedure. The input file with image data should contain
no metadata or some additional fields. It is regarded as a raw view of pixels.
The software generates a compressed file. Alongside with the compressed file,
 Springer Nature 2021 LATEX template

Combined Run-Length and Huffman Encoding for Image Compression 11

Fig. 6 Excerpt of the text file with distribution data

the software also stores individual compressed streams of color components so

the contribution of each element can be evaluated. For the test session, a set
of files was prepared to perform compression and decompression. The results
of the tests are aggregated in the Table 1.
As can be noticed from the presented results, the compression effectiveness
depends on the number of long sequences of the value 128. However, the role
of the Huffman compression cannot be neglected. It also can be observed that
the narrower histogram of values is, the better compression can be achieved.
Since the selected set of images are different by their structure, the results also
variate significantly. The best results were achieved for the gradient image (ID
- 6), while images that contain nature and comprise multiple details remain
in the close range (e.g., IDs 2, 3). The table actually established step-by-step
results of the encoding procedure. First, the overall length of the sequences that
are to be compressed is shown. Second, compressed sequences size is presented
alongside the compression ratio. Then, the size of the all data with RLE part
is demonstrated. It is followed by the results for the Huffman encoding. And,
finally, overall results are enlisted. Basically, using this table it is easy to follow
each step of the encoding process.
The coefficient column indicates the reduction coefficient that was applied
to obtain the file. Values of 1 and 3 were used during the test session.
As for the implementation of the approach, the value that is processed by
the RLE procedure is specified using the define directive. It can be changed
from the source code (recompilation of the source code is required in this case).
 Springer Nature 2021 LATEX template

12 Combined Run-Length and Huffman Encoding for Image Compression

Hence, the software is configurable and can be applied in other scenarios as a

tool for investigation or as a part of the other system.
One peculiarity to mention about the software implementation is that it
does not store the length in the file. However, as it launches the decoding
process immediately and regarding length field contribution as small, this can
be neglected. If necessary it can be easily updated to perform this action.
In general, it can be stated that the proposed approach demonstrated sig-
nificant compression results on the selected set of prepared images. Moreover,
providing that the input data follows the prescribed histogram pattern and
has one dominant value, the application of the method can be extended to
other images and data structures. On the other hand, it is not guaranteed that
the method works well on any data and can actually cause an increase of the
output file size. Hence, it should be applied only on the deliberately selected
and prepared input files.

5 Conclusions
The combined RLE-Huffman compression approach has been presented in the
paper. It is designed to work on the images that are prepared in the correspond-
ing way and satisfy some preliminary requirements. The proposed approach
relies on the sequential use of RLE and Huffman compression algorithms. The
described model was implemented in the software that was used further dur-
ing the test scenario. The results of the performed tests demonstrated that the
proposed technique is feasible and can ensure significant compression levels for
the specific data structures.

6 Declarations
6.1 Ethical approval
Not applicable.

6.2 Competing interests

The author has no competing interests to declare that are relevant to the
content of this article.

6.3 Authors’ contribution

Not applicable.

6.4 Funding
Not applicable.
 Springer Nature 2021 LATEX template

Combined Run-Length and Huffman Encoding for Image Compression 13

6.5 Availability of data and materials

The datasets generated during and/or analyzed during the current study are
available from the corresponding author on reasonable request.

References
Ahmed N, Natarajan T, Rao KR (1974) Discrete cosine transform. IEEE
transactions on Computers 100(1):90–93

Apostolico A (2007) Fast gapped variants for lempel–ziv–welch compression.

Information and Computation 205(7):1012–1026. https://doi.org/10.1016/
j.ic.2007.03.001, URL https://doi.org/10.1016/j.ic.2007.03.001

Elakkiya S, Thivya KS (2021) Comprehensive review on lossy and lossless com-

pression techniques. Journal of The Institution of Engineers (India): Series B
103(3):1003–1012. https://doi.org/10.1007/s40031-021-00686-3, URL https:
//doi.org/10.1007/s40031-021-00686-3

Facebook (2022) zst-fast real-time compression algorithm. URL https://

github.com/facebook/zstd, accessed: 2022-08-20

Knuth DE (1985) Dynamic huffman coding. Journal of algorithms 6(2):163–

180

Krainyk Y (2022) Compresssequence. URL https://github.com/codebreaker7/

CompressSequences, accessed: 2022-08-20

Kwan C, Larkin J, Budavari B, et al (2019) A short review of well-known

image codecs and observations and root cause of some strange behaviors in
image compression codecs. Signal, Image and Video Processing 14(3):583–
591. https://doi.org/10.1007/s11760-019-01591-5, URL https://doi.org/10.
1007/s11760-019-01591-5

Lee SW, Kim HY (2018) An energy-efficient low-memory image compression

system for multimedia IoT products. EURASIP Journal on Image and Video
Processing 2018(1). https://doi.org/10.1186/s13640-018-0333-3, URL https:
//doi.org/10.1186/s13640-018-0333-3

Nauman A, Qadri YA, Amjad M, et al (2020) Multimedia internet of things: A

comprehensive survey. IEEE Access 8:8202–8250. https://doi.org/10.1109/
access.2020.2964280, URL https://doi.org/10.1109/access.2020.2964280

Sayood K (2017) Introduction to data compression. Morgan Kaufmann

Schwarz S, Preda M, Baroncini V, et al (2018) Emerging mpeg standards for

point cloud compression. IEEE Journal on Emerging and Selected Topics in
Circuits and Systems 9(1):133–148
 Springer Nature 2021 LATEX template

14 Combined Run-Length and Huffman Encoding for Image Compression

Stabno M, Wrembel R (2009) RLH: Bitmap compression technique based

on run-length and huffman encoding. Information Systems 34(4-5):400–414.
https://doi.org/10.1016/j.is.2008.11.002, URL https://doi.org/10.1016/j.is.
2008.11.002

Wallace GK (1992) The jpeg still picture compression standard. IEEE

transactions on consumer electronics 38(1):xviii–xxxiv

Wang S (2019) Multimedia data compression storage of sensor network

based on improved huffman coding algorithm in cloud. Multimedia
Tools and Applications 79(47-48):35,369–35,382. https://doi.org/10.1007/
s11042-019-07765-0, URL https://doi.org/10.1007/s11042-019-07765-0

ZainEldin H, Elhosseini MA, Ali HA (2015) Image compression algorithms

in wireless multimedia sensor networks: A survey. Ain Shams Engineer-
ing Journal 6(2):481–490. https://doi.org/10.1016/j.asej.2014.11.001, URL
https://doi.org/10.1016/j.asej.2014.11.001
Table 1 Results of the proposed compression approach use on the set of prepared files

Image ID Coeff. T, bytes T128 , bytes T128compr Ratio128 Taf terRLE Taf terHuf f Ratiooverall

1 1 7372800 5731046 150212 38,15 1791966 808844 7,68

1 3 7372800 6388388 120664 52,94 1105076 335308 16,16
2 1 8294400 1396564 688084 2,02 7585920 2958087 2,27
2 3 8294400 5778900 465756 12,40 2981256 909854 6,02
3 1 8248320 248632 153184 1,62 8152872 3419848 2,30
3 3 8248320 4535823 1191420 3,80 4903917 1125317 3,56
4 1 16289280 636280 385100 1,65 16038100 6345080 2,42
4 3 16289280 9698591 2326212 4,16 8916901 1863882 3,88
5 1 16289280 4027531 1597508 2,52 14204857 4232314 2,85
5 3 16289280 13873418 1092092 12,70 3853554 763889 8,96
6 1 6220800 6189480 23764 260,45 55084 4983 216,39
6 3 6220800 6220800 12 518400 12 258 23040
7 1 7211520 339629 209736 1,61 7081627 2245938 2,93
7 3 7211520 5658502 1193572 4,74 2746590 390457 4,55
8 1 8294400 6084650 687384 8,85 2897134 836890 5,44
8 3 8294400 7778340 119216 65,24 635276 213645 24,91
9 1 7372800 5494441 619384 8,87 2497743 546360 6,32
9 3 7372800 7165600 70760 101,26 277960 207200 58,83
Springer Nature 2021 LATEX template

10 1 7372800 4920869 624192 7,88 3076123 811195 5,13

10 3 7372800 6863916 144124 47,62 653008 143588 25,62
11 1 7349760 1583409 514564 3,07 6280915 5766351 2,30
11 3 7349760 4670397 453472 10,29 3132835 962581 5,19
12 1 8294400 4930738 961804 5,12 4325466 1211552 3,81
Combined Run-Length and Huffman Encoding for Image Compression

12 3 8294400 7490742 183544 40,81 987202 266292 18,43

13 1 7395840 5047941 958996 5,26 3306895 646122 4,60
13 3 7395840 7186488 61260 117,31 270612 63413 59,32
15