World scientific research journal
https://scientific-jl.com/wsrj
Volume-38_Issue-1_April-2025
112
SSIM CRITERIA FOR ASSESSING IMAGE QUALITY
Jumayev Turdali Saminjonovich
PhD at the Department of “Modern information and communication
technologies”, International Islamic Academy of Uzbekistan
turdali240483@gmail.com
Tuhtanazarov Dilmurod Solijonovich
PhD at the Department of “Modern information and communication
technologies”, International Islamic Academy of Uzbekistan
dtuxtanazarov@gmail.com
Abstract.
This article reviews the theoretical and practical aspects of the SSIM
(Structural Similarity Index) criterion used in image quality assessment. SSIM is a
metric that determines the degree of similarity in images based on the human visual
system, allowing you to evaluate the differences between the structural structure,
brightness and contrast of an image. Compared to traditional criteria such as MSE
(Mean Squared Error) and PSNR (Peak Signal-to-Noise Ratio), SSIM evaluates
image quality closer to real viewing conditions. The article analyzes in detail the
SSIM formula, its main components and calculation methods. It also discusses the
practical application of SSIM, its advantages and limitations, and highlights its role
in image quality assessment.
Keywords.
SSIM, image quality assessment, structural similarity index, PSNR,
MSE, image analysis, noise effect, contrast, brightness, visual quality, computer
vision, compressed images, algorithmic evaluation.
INTRODUCTION
Image quality assessment plays an important role in the processing,
compression, and restoration of digital images. Various algorithms and methods are
used to assess the sharpness, visual quality, and similarity of images. Traditional
methods, including MSE (Mean Squared Error) and PSNR (Peak Signal-to-Noise
Ratio), mathematically express quality by calculating pixel-level differences in
images. However, these methods cannot fully reflect the complex characteristics of
the human visual system.
SSIM (Structural Similarity Index) takes a more precise approach to assessing
image quality by taking into account how the human eye perceives images. This index
evaluates the quality by taking into account important factors such as brightness,
contrast, and structural similarity of images. Therefore, SSIM is widely used in many
fields, including medical imaging, video compression, artificial intelligence, and
computer vision technologies.
World scientific research journal
https://scientific-jl.com/wsrj
Volume-38_Issue-1_April-2025
113
This article reviews the working principle of the SSIM criterion, its
mathematical model, and comparison with other quality assessment methods. The
practical application, advantages, and limitations of SSIM are also analyzed.
Problem statement.
Suppose we are given a set of
𝑛
biometric images:
𝑇
1
, … , 𝑇
𝑖
, … , 𝑇
𝑛
,
where:
𝑛
is the number of images,
𝑇
𝑖
is the given
𝑖
-th biometric image.
The main goal is to develop a criterion for assessing the quality of the given
biometric images.
Method of solving the problem.
By using SSIM (Structural Similarity Index)
to assess image quality, we can more accurately assess the true quality of images from
the perspective of the human visual system. The solution to the problem involves the
following steps:
1.
Analysis of the SSIM formula and its components
SSIM is a measure of image quality based on brightness, contrast, and structural
similarity. The following formulas are used to calculate this indicator:
•
Luminance
: Average brightness values are compared to evaluate the
brightness of images.
•
Contrast
: Average contrast values are calculated to compare the contrast of
images.
•
Structure
: Image structure is compared to determine structural similarities
between images.
The SSIM formula is expressed as follows:
𝑆𝑆𝐼𝑀 =
(2𝜇
𝑥
𝜇
𝑦
+ 𝐶
1
)(2𝜎
𝑥𝑦
+ 𝐶
2
)
(𝜇
𝑥
2
+ 𝜇
𝑦
2
+ 𝐶
1
)(𝜎
𝑥
2
+ 𝜎
𝑦
2
+ 𝐶
2
)
where:
𝜇
𝑥
, 𝜇
𝑦
- average brightnesses of images;
𝜎
𝑥
2
, 𝜎
𝑦
2
- contrasts of images.
𝜎
𝑥𝑦
- covariance between images.
𝐶
1
, 𝐶
2
- small constants, which are used to reduce wall noise.
2. Comparing Images
To calculate SSIM, you need two images: the original (or high-quality) image
and the compressed or processed image. Divide each image into small blocks and
compare each of them using the SSIM formula above. An SSIM value is calculated
for each block and these values are combined. As a result, the overall SSIM value
gives the overall similarity between the images.
3. Comparing SSIM and other metrics
Traditional evaluation methods, such as PSNR and MSE, are compared with
SSIM. While these metrics only measure pixel-level differences, SSIM takes into
World scientific research journal
https://scientific-jl.com/wsrj
Volume-38_Issue-1_April-2025
114
account the structure, brightness, and contrast of the images. PSNR and MSE often
show differences between images with high values, but they do not reflect significant
differences for the human visual system. SSIM, on the other hand, better reflects the
visual quality of the images.
4. Application of SSIM in real practice
When using SSIM, it is possible to test various aspects of images, for example,
by changing the compression level, checking the image quality in video transmission
processes, or assessing the similarity in medical images. The application of SSIM in
real practice more accurately shows the visual quality of the image and allows for a
more realistic assessment of quality.
5. Analysis of the results
In the final stage, the SSIM values are compared with the PSNR and MSE
values. A high value of SSIM indicates that the image is close to the original, which
means that the image quality is good. The results are analyzed and the effectiveness
of SSIM in determining the quality of different levels of images is demonstrated.
With this method, it is possible to study the effectiveness of the SSIM criterion
and accurately and realistically assess image quality. This allows for higher quality in
image compression and processing processes.
Experimental results.
In this study, the effectiveness of the SSIM (Structural Similarity Index) criterion
for assessing image quality was compared with traditional methods, such as PSNR
and MSE. Several different images were studied in the study, including compressed
and processed images, as well as original (high-quality) images. The results of the
study were analyzed based on the following main conclusions:
1. SSIM values and image quality assessment
In the study, SSIM values were higher than 0.9 for clear and high-quality images.
This indicated that SSIM was close to the original image, i.e., high SSIM values
minimized the structures and visual differences between images. However, in
compressed or processed images, SSIM values decreased, indicating a loss of image
quality compared to the original.
2. Comparison of SSIM and PSNR
PSNR values may be higher than SSIM, but PSNR only takes into account pixel-
level differences, which does not fully reflect how an image appears to the human
visual system. For example, SSIM produced more accurate results when assessing
visual similarity between images. In some cases, a higher PSNR value did not indicate
a deterioration in the original visual quality or an increase in noise in the images.
3. Comparison of MSE and SSIM
When MSE values were compared with SSIM values, SSIM reflected the
sensitivity of the human visual system more. MSE values reduced the overall quality
of the images by measuring only pixel-level differences. SSIM, on the other hand,
World scientific research journal
https://scientific-jl.com/wsrj
Volume-38_Issue-1_April-2025
115
reflected the visual quality of the images better by focusing on the structures and
details of the images.
4. Practical examples with SSIM
The study applied the SSIM criterion to various real-world applications, such as
image compression, video transmission, and similarity assessment in medical images.
For example, in medical image processing, SSIM showed high results and allowed
doctors to detect subtle differences between images, which is important for doctors.
In compressed images, SSIM values decreased, indicating a loss of image integrity
and sharpness.
Figure 1. Measuring image quality based on the SSIM metric
Conclusion.
An SSIM (Structural Similarity Index) value of 1 indicates that the
full structures and visual properties of the two images are perfectly preserved. This
indicates that there is no noticeable difference between the original and the
compressed or modified images. This means that all the main elements of the images,
including structural details, color contrast, texture, and unchanging structure, are
preserved.
The similarity between the images is not only in the overall colors and
brightness, but also in the structure and structural details of the image. This means
that the lines, shapes, and geometric changes in the image are completely consistent.
Usually, image compression or transmission processes change the structures in the
image, but when the SSIM value is 1, there is no noticeable negative effect from these
processes.
An SSIM value of 1 confirms the high quality of the image, because all the
properties of the image, such as contrast, details, colors, and other visual elements,
are preserved as in the original state. This means that the image has not been adversely
affected by the compression or transformation processes.
Compression processes usually cause some loss of quality in the image, such as
loss of detail or changes in common colors and textures. However, an SSIM of 1
indicates that there are no such changes in the compressed or transformed image.
When compressing or transforming an image, it is usually possible to lose some
detail or changes that are visually noticeable. For example, small edges in the image,
World scientific research journal
https://scientific-jl.com/wsrj
Volume-38_Issue-1_April-2025
116
certain differences in black or white, or structural details. However, an SSIM value
of 1 means that no noticeable difference or loss has occurred in this process.
References
:
1. Saminjonovich, T. J. (2022). Algorithm for extraction of identification
features in ear recognition.
ACADEMICIA: An International Multidisciplinary
Research Journal
,
12
(1), 232-237.
2. Fazilov, S. X., Mahkamov, A. A., & Jumayev, T. S. (2018). Algorithm for
extraction of identification features in ear recognition. In
Информатика: проблемы,
методология, технологии
(pp. 3-7).
3. Тoirov, B. Т., Jumaev, Т. S., & Toirov, O. T. (2021). Obyektlarni tanib olishda
python dasturidan foydalanishning afzalliklari.
Scientific progress
,
2
(7), 165-168.
4. Saminjonovich, J. T., & Solijonovich, T. D. (2023). Algorithm for improving
the quality of image for the person identification. Galaxy International
Interdisciplinary Research Journal, 11(9), 372-375.
5. FazilovSh.X., MahkamovA.A., JumayevT.S. Algorithm for extraction of
identification features in ear recognition // Informatics: problems, methodology,
technologies: Materials of the XVII international scientific and methodological
conference. Voronezh, 2018.Vol. 2 - p. 3-7. p.
6. Mirzaev O.N., Djumaev T.S. Construction of personality recognition
algorithms based on the image of the ears // Current state and prospects for the use of
information technologies in management: Reports of the Republican Scientific and
Technical Conference. September 7-8, 2015. CRPP and agro-industrial complex at
TUIT, 342-348 p.
7. Jumaev T.S. Algorithm for distinguishing the characteristics of the image of
the ear on the basis of discrete cosine displacement // TATU messages. - Tashkent,
2011. -№2. 74-78 p.
8. Mirzaev, N.M., Radjabov, S.S., & Djumaev, T.S. (2008). On the
parameterization of models of recognition algorithms based on the assessment of the
interrelation of features. Informatics and energy problems, (2-3), 23-27.
9. Jumayev, T. S., Mirzayev, N. S., & Makhkamov, A. S. (2015). Algorithms for
segmentation of color images based on the allocation of strongly coupled elements.
Studies of technical sciences, (4), 22-27.
10. Mirzayev, N. M., S. S. Radjabov, and T. S. Zhumayev. "O parametrizatsii
modeley algoritmov raspoznavaniya, osnovannyh na otsenke vzaimosvyazannosti
priznakov." Problemy informatiki i energetiki (2008): 2-3.
11. Mirzayev N. M., Radjabov S. S., Jumaev T. S. Isolation of characteristic
features of facial images in personality recognition problems. Neurocomputers and
their application.-2016.
12. Фазылов Ш. Х., Мирзаев Н. М., Махкамов А. А. Выделение
геометрических признаков изображений ушных раковин //XI всероссийская
научная конференция «нейрокомпьютеры и их применение. – 2013. – Т. 19.
13. Jumayev Turdali Saminjonovich, & Mahkamov Anvarjon Abdujabborovich.
(2021). Algorithm for extraction of identification features in ear recognition.
International Journal of Innovations in Engineering Research and Technology, 7(05),
216–220.
