A Novel Approach for Enhancing Contrast for Digital

Images
Dr Umesha K Dept. of Electronics
and communication Engg,Jawaharlal
College of Engineering and Technology

Abstract— Enhancing contrast is essential for improving the and color representation. There are two types of CE
visual appeal and automatic decision-making in industrial techniques: direct and indirect [1]. Image contrast is
applications across a variety of disciplines. This work provides a evaluated using direct methodologies based on the human
unique method based on Stevens' power law (SPL), which is
visual system (HVS), such as the Weber Fechner law or
derived from human brightness perception, for changing the
tonality of many types of images, including grayscale or color
Retinex theory [2], and is improved by employing various
contrast-distorted images and medical images. The suggested nonlinear functions [3], or addressing optimization problems
approach takes into account the non-linear relationship between [4]. Direct techniques have some advantages in terms of
intensity and perception in order to improve the overall tonal look image detail enhancement and dynamic range reduction, but
of the image. In order to attain the correct tone and improve visual they require a high computational complexity and produce
appeal, the first phase includes tonal adjustment utilizing SPL, 'halo' aberrations, particularly around sharp edges [5].
which enables exact fine-tuning of brightness levels depending on Despite the fact that new direct approaches have been
intensity values. A sigmoid function is used to improve contrast by developed to address these challenges, it remains
selectively amplifying changes in pixel brightness between
challenging to provide both high picture contrast and real-
neighboring pixels. This improves overall contrast while keeping
important details and avoiding over-amplification. The suggested time processing without introducing noticeable distortion. In
approach has two main advantages: it provides optimal visibility practice, indirect procedures, based on a global
and it has lower computational complexity. The processing transformation function, are more typically utilized than
demands are reduced without affecting the output's quality by direct methods for these reasons: when there are high peak
utilizing SPL and the sigmoid function. This makes the suggested values in the image histogram, one of the most commonly
solution suitable for real-time image processing applications while used indirect procedures, histogram equalization (HE),
also making it effective. Results from experiments show how the produces visual distortions such as noise amplification,
suggested approach may be used to improve the visual quality of contouring, or considerable brightness change [6]. Several
many types of photos by altering the tone, adding contrast,
eliminating processing artifacts, and more. Performance
HE-based techniques solve these difficulties by modifying
comparisons with other algorithms demonstrate the method's better the image histogram to lower high peak values before
performance and efficiency. deriving the transformation function from the adjusted
histogram. Recent indirect ways for preserving detail
Keywords: Steven’s power law, contrast enhancement, construct the picture histogram using a two-dimensional
contrast stretching. (2D) histogram [7] or fuzzy contextual information [8],
which provides higher weight to pixels in texture regions.
I. INTRODUCTION
When compared to earlier indirect methods, these methods
A key visual characteristic that is essential to how we see improve performance and reduce information loss.
and interpret pictures is contrast. It describes the variation in Throughout the years, a plethora of algorithms have been
brightness, color, or intensity between the different introduced on CE, ranging from simple to complicated in
components of a picture. To successfully communicate terms of calculations and concepts. In their study, Singh et
visual information and improve the visual quality of al. [9] provided a method termed recursive exposure sub
pictures, there must be sufficient contrast. To increase the image histogram equalization (RESIHE). RESIHE
visual quality of the photos, researchers have proposed iteratively performs enhanced sub-picture histogram
different enhancement strategies. Contrast enhancement equalization (ESIHE) to a picture until succeeding rounds of
(CE) is a technique for improving the appearance of images exposure change are less than a predefined threshold. Jiang
in order to make them more suitable to human sight. CE is et al. [10] created an optimal gamma correction and
usually made up of a series of operations, such as increasing weighted sum (OGCWS) algorithm that aims to improve
the disparities between foreground and background objects. contrast while retaining the average. In order to increase the
Since, CE methods can boost an image's brightness and/or color vividness of an image, Wong et al. [11] developed the
contrast to amplify features and make them easier to see, the histogram equalization and optimal profile compression
enhancement field is significant for both computer vision (HEOPC) technique. Then, after converting the primary
and human perception. CE is widely used in a variety of colors into linked human perceptual zones, it modifies the
applications and industries, including medical imaging, image's intensity using a newer HE method that compresses
video surveillance systems, satellite image processing, unwanted artefacts. A fuzzy contextual (FC) technique for
remote sensing, biology, and geographic research. Many CE color and grayscale pictures was introduced by Parihar et al.
algorithms try to improve image quality attributes as [12]. The chosen algorithm amplifies intensity variations
consistent contrast, brightness preservation, noise tolerance, based on the precise image's local relative information. The
use of a fuzzy dissimilarity histogram (FDH) allows for the an algorithm should make an effort to reduce processing
incorporation of contextual data. Convolutional neural mistakes and prevent the introduction of undesirable
networks (CNNs), in particular, have been the subject of artifacts. A unique technique can significantly increase CE
substantial research for picture improvement in more recent by resolving these issues, resulting in improved visual
years. To the best of our knowledge, however, the majority quality and correct perception of picture details.
of deep learning-based techniques have been created for In order to achieve the aforementioned goals, "Dynamic
medical photos [13], low-light images [14], or hazy images Contrast Enhancement using Power Sigmoid" is proposed.
[15] rather than for natural image augmentation. A quick
and effective image contrast enhancement algorithm III. BACKGROUND
(HLIPSCS) was proposed by Zohair et al [16]. Better A. Human Brightness Perception
brightness preservation, enhanced contrast, and more vibrant
The brain's interpretation of the various light wavelengths
colors were the results of this algorithm, using many
picked up by the cone cells in the retina is what allows
concepts like statistics, hyperbolic functions, contrast
humans to see color. Cone cells, which are sensitive to the
stretching, and logarithmic image processing.
three fundamental colors red, green, and blue, are in charge
of color vision. In order to perceive a variety of colors, the
II. RELATED LITERATURE brain analyses the data from these cones. Color spaces, like
the RGB (Red-Green-Blue) color space, can be used to
In comparison to other contrast enhancement techniques, explain how colors are perceived. Each color is represented
such as [9], [10], [11], [12], Zohair et al.'s method[16] by a combination of red, green, and blue intensities in the
offers: (a) efficient processing for a variety of grayscale and RGB color model. In an 8-bit color depth system, intensities
color pictures with poor contrast representations, (b) vary from 0 to 255, with higher numbers signifying higher
produce the output quickly while avoiding processing errors, intensities.
and (c) provide an enhancement amount control parameter The brain uses the variations in pixel intensities to detect
in order to make it simpler for the operator to get the desired changes in intensity in an image. Mathematical formulas and
outcomes. models can be used to explain how the brain reacts to
The following straightforward explanation will show how variations in intensity. Weber's law [17], which asserts that
their approach functions: firstly, two separate hyperbolic the perceived change in intensity is proportionate to the
functions process the input picture. Next, a LIP model is starting intensity, is one widely used model. It can be written
used to integrate the output of these functions. The contrast mathematically as:
of the picture is then improved by using a nonlinear
transformation function in the form of a sigmoid function.
dI
After that, a statistical processing technique is used to =k (1)
enhance the image's brightness. And last, normalization, I
which is the final algorithm output, does contrast stretching.
Better brightness preservation, enhanced contrast, and more where ‘k’ is a constant known as Weber's fraction, I stands
vibrant colors are the results of this algorithm, using many for the original intensity, dI stands for the intensity change.
concepts like statistics, hyperbolic functions, contrast The sensitivity of the human brain to intensity fluctuations is
stretching, and logarithmic image processing. determined by Weber's fraction. A lower value of ‘k’
In their method, hyperbolic functions were added to denotes increased sensitivity to intensity variations.
enhance an image's tonality, but these functions induce a fast The idea of brightness is a key one in the perception of
rise in pixel value, which in turn leads to a compression of color. Luminance, which is obtained from the RGB color
the tonal range overall and a loss of details. Additionally, values, stands for the brightness or intensity of a picture.
hyperbolic functions do not offer precise control over the B. Human Perception Based Steven’s Power Law
tone mapping procedure. Due to the mathematical definition
of these functions, they do not provide modifiable Based on SPL [18], it is hypothesized that human perception
parameters that can be easily modified to achieve desired has a non-linear connection with physical stimulus intensity
tonal alteration. Additionally, it can be computationally for a variety of sensory inputs, such as brightness, loudness,
expensive to calculate hyperbolic functions for larger and perceived magnitude. This connection is described
photos. In time-sensitive applications or real-time scenarios, mathematically by SPL.
this complexity may result in an increase in the processing The physical intensity (I) increased to a power (n)
time needed for tone augmentation. In addition to hyperbolic determines the perceived magnitude (P) of a stimulus, in
functions, their method combines hyperbolic tangent and accordance with SPL. It has the following mathematical
sine characteristics with LIP models to produce distinctive expression:
images. LIP model performance significantly depends on the
n
calibre , variety, and representation of training data. LIP P=K∗I (2)
models are used to combine hyperbolic sine and tangent
features, which involves complex calculations needing a lot In this equation, ‘n’ is a parameter that controls how steep
of computational power and processing time. the connection is, and ‘K’ is a constant. The sensitivity of
Nevertheless, despite improvements in CE methods, perception to variations in physical intensity is indicated by
there is still room for an algorithm to be developed that the value of ‘n’. Small increases in intensity translate into
performs better in terms of simplicity, effectiveness, and bigger apparent changes in magnitude when ‘n’ is larger,
maintaining both brightness and color representation. Such indicating a steeper connection. A lower number of ‘n’, on
the other hand, denotes a shallower connection, with smaller 1
perceived changes in magnitude for a given change in SF= −P
intensity. 1+ e
According to SPL, the physical strength of a stimulus (4)
does not necessarily rise linearly with our perception of it.
Instead, depending on the magnitude of n, it has a The sigmoid function maps the pixel values using a non-
compressive or expanding impact. These properties of our linear mapping, compressing the intensity range and
sensory systems, which adapt and react differently to varied extending the values towards the extremes.
intensities, are reflected in this nonlinear connection. In CE process, brightness of image should be improved.
For instance, SPL suggests that our perception of Here an improved version of CDF of Gompertz distribution
brightness does not rise in direct proportion to the physical [21] is used in order to increase the brightness which is
luminance in the context of brightness perception. The utilized as :
perception of modest increases in brightness at lower
{−( 0.4∗d )∗(e SF −1)} (5)
intensities is stronger than that of small increases at higher G=1−e
intensities. Our visual system can successfully adjust to a
broad variety of brightness levels thanks to this non-linear The ‘d’ parameter controls the enhancement effect, with
connection, emphasizing features in darker areas while larger values resulting in more pronounced enhancement.
retaining discriminating in brighter areas. Finally, the generated image G is adjusted for intensity by
Understanding human perception in terms of SPL can using method called normalization, which scales the
help us better understand how our sensory systems react to intensity values to the full dynamic range. The
stimuli and can also help us create perceptually correct normalization function is given as [22]:
methods for processing images and sounds. We may create
algorithms and models that take into account these perceptual
qualities by taking into account the non-linear structure of CE=
[ G−min ( G ) ]
perception, which will eventually result in more accurate and [ max ( G )−min (G ) ]
useful sensory experiences. (6)
IV. PROPOSED WORK
The main reasons behind introducing SPL in proposed
method are: (a) SPL seeks to preserve the image's perceptual
linearity, which means that variations in pixel values more
closely match how people perceive things. This provides a
more attractive and realistic-looking image, (b) the exponent
parameter of SPL enables precise control of the tone
mapping procedure. This parameter allows for more
freedom in tone modifications by allowing the
transformation to be tailored to certain image attributes or
artistic preferences, and (c) compared to the hyperbolic
functions, the power law transformation maintains more
picture details. This implies that significant details and
minute differences in the picture may be more effectively Fig. 1. Workflow diagram
preserved, leading to a more accurate depiction of the
This adjustment guarantees that the improved image,
original image.
denoted by CE, makes full use of the available pixel value
Going into specific details, consider an input image,
range, maximizing the CE impact and delivering a visually
which may be medical photos, satellite images, underwater
appealing outcome. Fig. 1 shows the workflow diagram of
images, or images with altered contrast. To begin, apply
proposed method.
SPL transformation to the supplied image. Each pixel value
in the image is increased to the power of ‘k’, resulting in a
transformed image given by: V. EXPERIMENTAL RESULTS
k A number of tests are conducted to show how well the
P=I (3) suggested algorithm works. Thousands of pictures were
This non-linear adjustment modifies the image's downloaded from the internet and hundreds from Berkley
intensities, emphasizing higher values and potentially image data set [25], some of which were natural and others
increasing contrast. The contrast of the idealized image is of which had true contrast distortion. Some of these
enhanced even further. The converted picture P is then put photographs were utilized for comparison with current
through a sigmoid function (SF) [19] ,which is the S-curve techniques while others were used for experimental reasons.
transformation function utilized to enhance contrast in To compare the perceived quality of the outcomes obtained
several research articles[20]. The SF is defined as: from the suggested and comparable approaches, two picture
assessment methodologies are used. These techniques
include Spatial Frequency (SF) [23] and the Natural Image
Quality Evaluator (NIQE) [24]. More crucially, the
suggested method's and the comparison method's execution
times are compared. Assessing and quantifying the (e1) (f1) (g1) (h1)
distribution of the numerous spatial frequencies seen in an
image is the process of spatial frequency assessment in
images. It involves looking at the patterns and variations in
color or intensity at different scales or frequencies.
In order to capture several statistical aspects of the (e2) (f2) (g2) (h2
picture, such as gradients, local brightness statistics, and
color attributes, the NIQE method computes a collection of
features. The quality score of the image is then estimated
using these features. The perceived quality of the image .
decreases as the NIQE score increases. (e3) (f3) (g3) (h3)

Fig. 3. Real color images from Berkley image data set [25] processed by
[16] and proposed method .

In comparison to the approach given in [16], the

experimental results shown in Fig. 2 and Fig. 3 demonstrate
(a1) (b1) (c1) (d1)
how well our suggested method for CE performs. While ‘a2’
to ‘h2’ correspond to the contrast-enhanced versions of ‘a1’
to ‘h1’ using the approach in [16], ‘a1’ to ‘h1’ represent the
input pictures. The contrast-enhanced pictures of ‘a1’ to ‘h1’
that were acquired using our suggested approach are
represented similarly by ‘a3’ to ‘h3’.
(a2) (b2) (c2) (d2) It is clear from the findings that our suggested strategy
efficiently improves the tonality of the photos by adjusting
the SPL's parameter "k". It is clear from comparing
assessment metrics like NIQE, SF, and processing time that
our technique outperforms [16] in terms of visual quality
and processing speed. Notably, the suggested technique
(a3) (b3) (c3) (d3) delivers more vibrant colors while effectively maintaining
the original brightness of color photographs, resulting in a
Fig. 2. Real contrast distorted images processed by [16] and proposed
method. natural look.

NIQE has the benefit of not requiring a reference picture,

which makes it appropriate for assessing image quality when TABLE 1. The scored image evaluation readings and execution times of
two compared algorithms.
one is not readily available or practicable to produce. In
real-time applications including video processing, security
systems, and medical imaging, CE techniques are frequently Method [16] Proposed method
utilized. For the system to operate in real-time or to maintain
a smooth user experience, these need rapid and responsive IMAGE NIQE SF TIME NIQE SF TIME
processing. Making ensuring the CE procedure can be
Fig. a1 5.225 0.66 0.176 5.274 0.67 0.154
completed within the appropriate time restrictions involves
evaluating execution time. Fig. b1 3.964 0.61 0.033 4.130 0.63 0.027
We used a computer system with a 2.8 GHz Intel Core i7
CPU and 16 GB of RAM to conduct the testing and Fig. c1 2.724 0.62 0.017 2.767 0.62 0.008
comparisons. The environment used to execute all the Fig. d1 2.342 0.64 0.024 2.368 0.66 0.017
programmes is MATLAB 2021a.
Fig. e1 3.359 0.57 0.177 3.409 0.58 0.138
Fig. 2 shows the experimental results of various real
Fig. f1 2.490 0.60 0.027 2.795 0.59 0.012
contrast distorted images and some medical images.
Different images are processed separately to generate the Fig. g1 2.528 0.58 0.036 2.589 0.58 0.024
perfect tone by altering the constant parameter "k" of SPL.
This gives the hue mapping action fine-grained control and Fig. h1 3.232 0.63 0.035 3.246 0.65 0.012
caters to the distinctive features and needs of each image. Avg 3.233 0.61 0.065 3.322 0.622 0.049
The hue may be successfully improved by choosing the right
value of "k" for a given image, leading to aesthetically The data from two picture assessment approaches are shown
stunning effects that go beyond the capabilities of in Table 1, which offers thorough insights into the
hyperbolic functions. effectiveness of several strategies, including our suggested
method. Notably, our approach performs better than others
in terms of effectiveness and processing speed. The
suggested method's average NIQE and SF values are greater
than those of [16], demonstrating our method's superior
performance in contrast augmentation. Our suggested [3] S. C. Nercessian, K. A. Panetta, and S. S. Agaian, ``Non-linear direct
solution also runs faster than [16], demonstrating its multiscale image enhancement based on the luminance and contrast
masking characteristics of the human visual system,'' IEEE Trans.
quickness and effectiveness among all. Image Process., vol. 22, no. 9, pp. 3549_3561, Sep. 2013
Additionally, Table 2 compares the experimental outcomes [4] H. Yue, J. Yang, X. Sun, F.Wu, and C. Hou, ``Contrast enhancement
for two authentically altered photos between our suggested based on intrinsic image decomposition,'' IEEE Trans. Image
approach and numerous previously described algorithms. Process., vol. 26, no. 8, pp. 3981_3994, Aug. 2017.K. Muhammad, R.
Hamza, J. Ahmad, J.Lloret, H.Wang, and S.W. Baik, “Secure
TABLE 2. Image evaluation readings of two measures: NIQE and surveillance framework for IoT systems using probabilistic image
SF for six methods. encryption,” IEEE Trans. Ind. Informat., vol. 14, pp. 3679–3689,
2018.
[5] T. Celik and T. Tjahjadi, ``Automatic image equalization and contrast
Method Image Parameters enhancement using Gaussian mixture modeling,'' IEEE Trans. Image
Pro- cess., vol. 21, no. 1, pp. 145_156, Jan. 2012.
[6] T. Arici, S. Dikbas, and Y. Altunbasak, ``A histogram modifcation
NIQE SF framework and its application for image contrast enhancement,'' IEEE
Trans Image Process., vol. 18, no. 9, pp. 1921_1935, Sep. 2009.
a1 4.787 0.677
[7] S.-W. Kim, B.-D. Choi, W.-J. Park, and S.-J. Ko, ``2D histogram
[09] equalization based on the human visual system,'' Electron. Lett., vol.
h1 3.427 0.344
52, no. 6, pp. 443_445, 2016
a1 4.980 0.701 [8] A. S. Parihar, O. P. Verma, and C. Khanna, “Fuzzy-contextual
contrast enhancement,” IEEE Transactions on Image Processing, vol.
[10] h1 3.289 0.390 26, no. 4, pp. 1810–1819, 2017.
[9] K. Singh, R. Kapoor, and S. K. Sinha, “Enhancement of low exposure
a1 4.859 0.682 images via recursive histogram equalization algorithms,” Optik, vol.
126, no. 20, pp. 2619–2625, 2015.
[11] h1 3.344 0.365 [10] G. Jiang, C. Wong, S. Lin, M. Rahman, T. Ren, N. Kwok, H. Shi, Y.-
H. Yu, and T. Wu, “Image contrast enhancement with brightness
a1 4.575 0.664 preservation using an optimal gamma correction and weighted sum
approach,” Journal of Modern Optics, vol. 62, no. 7, pp. 536–547,
[12] h1 3.469 0.310 2015.
[11] C. Y. Wong, G. Jiang, M. A. Rahman, S. Liu, S. C.-F. Lin, N. Kwok,
a1 5.025 0.661 H. Shi, Y.-H. Yu, and T. Wu, “Histogram equalization and optimal
profile compression based approach for colour image enhancement,”
[16] h1 3.359 0.575
Journal of Visual Communication and Image Representation, vol. 38,
pp. 802–813, 2016.
a1 5.274 0.679
[12] A. S. Parihar, O. P. Verma, and C. Khanna, “Fuzzy-contextual
PROPOSED h1 3.409 0.589 contrast enhancement,” IEEE Transactions on Image Processing, vol.
26, no. 4, pp. 1810–1819, 2017.
[13] M. Li, S. Shen, W. Gao, W. Hsu, and J. Cong,”Computed tomography
image enhancement using 3D convolutional neural network,'' in Deep
The results shown in the Table 2 show that our approach Learning in Medical Image Analysis and Multimodal Learning for
has the best NIQE and SF scores among all algorithms and Clinical Decision Support. Cham, Switzerland: Springer, 2018, pp.
the quickest execution time, further demonstrating its 291_299.
efficiency and effectiveness in comparison to other [14] W. Ren, S. Liu, L. Ma, Q. Xu, X. Xu, X. Cao, J. Du, and M.-H. Yang,
approaches. ``Low-light image enhancement via a deep hybrid network,'' IEEE
Trans. Image Process., vol. 28, no. 9, pp. 4364_4375, Sep. 2019.
VI . CONCLUSION [15] C. Hodges, M. Bennamoun, and H. Rahmani, ``Single image dehazing
using deep neural networks,'' Pattern Recognit. Lett., vol. 128, pp.
This work proposes an efficient image enhancement scheme 70_77, Dec. 2019.
for industrial applications. Compared to state of the art the [16] Al-Ameen, Zohair & Younis, Zainab & AL-Ameen, Shamil. .
suggested method based on SPL successfully increases HLIPSCS: A Rapid and Efficient Algorithm for Image Contrast
contrast, boosts visual appeal, and shortens processing time, Enhancement. 12. 311-320. 10.12785/ijcds/120125.Juky 2022
[17] S. Park, Y.-G. Shin, and S.-J. Ko, “Contrast enhancement using
with the possibility for future modifications to make the
sensitivity model-based sigmoid function,” IEEE Access, vol. 7, pp.
power law flexible. Additionally, the technique effectively 161 573–161 583, 2019.
maintains critical pixel brightness changes while selectively [18] S. S. Stevens, ``On the psychophysical law,'' Psychol. Rev., vol. 64,
magnifying changes in picture detail, resulting in better no. 3, pp. 153_181, 1957.
contrast and tonal appearance overall. The experimental [19] S. K. Swee, L. C. Chen, and T. S. Ching, “Contrast enhancement in
findings show the suggested approach's improved endoscopic images using fusion exposure histogram equalization.”
Engineering Letters, vol. 28, no. 3, 2020.
performance, effectiveness, and promise for a range of
[20] M. Wan, G. Gu, W. Qian, K. Ren, and Q. Chen, “Infrared small target
image processing applications like healthcare and industry enhancement: grey level mapping based on improved sigmoid
4.0, etc. transformation and saliency histogram,”. Journal of Modern Optics,
vol.65, no.10,pp.1161-1179,2018.
[21] Z. Al-Ameen, H. N. Saeed, and D. K. Saeed, “Fast and efficient
REFERENCES algorithm for contrast enhancement of color images,” Review of
[1] L. Dash and B. N. Chatterji, ``Adaptive contrast enhancement and Computer Engineering Studies, vol. 7, no. 3, pp. 60–65, 2020
deenhancement,'' Pattern Recognit., vol. 24, no. 4, pp. 289_302, 1991. [22] S. Park, Y.-G. Shin, and S.-J. Ko, “Contrast enhancement using
[2] E. H. Land and J. J. McCann, ``Lightness and Retinex theory,'' J. Opt. sensitivity model-based sigmoid function,” IEEE Access, vol. 7, pp.
Soc. Amer.,vol.61,no.,1,pp.1_11,1971 161 573–161 583, 2019.
[23] S. Li, J. T. Kwok, and Y. Wang, “Combination of images with diverse [24] A. Mittal, R. Soundararajan, and A. C. Bovik, “Making a “completely
focuses using the spatial frequency,” Information Fusion, vol. 2, no. 3, blind” image quality analyzer,” IEEE Signal Processing Letters, vol.
pp. 169–176, 2001. 20, no. 3, pp. 209–212, 2012.
[25] Berkley image data set. https://www2.eecs.berkeley.edu