129
3C Tecnología. Glosas de innovación aplicadas a la pyme. ISSN: 2254 – 4143 Edición Especial Special Issue Marzo 2020
ENHANCING UNDERWATER IMAGES USING PIECEWISE
LINEAR SMOOTHING GRADIENT GUIDED FILTER
A. Chrispin Jiji
Assistant Professor, Department of Electronics and Communication Engineering, The Oxford
College of Engineering, Bangalore-560068 and aliated to Visvesvaraya Technological
University, Belagavi, Karnataka, (India).
E-mail: chrispinjij@gmail.com ORCID: https://orcid.org/0000-0001-5267-788X
N. Ramrao
Vice Chancellor, Kalasalingam University, Srivilliputtur, Tamilnadu, (India).
E-mail: nagaraj.ramrao@gmail.com ORCID: https://orcid.org/0000-0003-2542-5999
Recepción:
05/12/2019
Aceptación:
30/12/2019
Publicación:
23/03/2020
Citación sugerida:
Chrispin Jiji, A., y Ramrao, N. (2020). Enhancing underwater images using piecewise linear smoothing
gradient guided lter. 3C Tecnología. Glosas de innovación aplicadas a la pyme. Edición Especial, Marzo 2020,
129-139. http://doi.org/10.17993/3ctecno.2020.specialissue4.129-139
Suggested citation:
Chrispin Jiji, A., & Ramrao, N. (2020). Enhancing underwater images using piecewise linear smoothing
gradient guided lter. 3C Tecnología. Glosas de innovación aplicadas a la pyme. Edición Especial, Marzo 2020,
129-139. http://doi.org/10.17993/3ctecno.2020.specialissue4.129-139
130 http://doi.org/10.17993/3ctecno.2020.specialissue4.129-139
3C Tecnología. Glosas de innovación aplicadas a la pyme. ISSN: 2254 – 4143 Edición Especial Special Issue Marzo 2020
ABSTRACT
Poor visibility owing towards illumination absorption and scattering is challenging
for processing undersea descriptions. However, for enhancing true scene from such
degradation is more important. Unfortunately, existing methods cause gradient reversal
artifact particularly near boundaries. To get better insight of undersea imagery, we project
a piecewise linear smoothing Gradient Guided Filter (P-GGF) technique is to defeat the
diculties caused by conventional schemes, hence produce sharper boundaries based on
GGF and smoothed output based on piecewise linear model. The projected technique
mainly functional for smoothing, ash and feathering. Tentative results prove that the
resultant algorithm can produce imagery with improved ocular excellence than existing
methods.
KEYWORDS
Image Enhancement, Guided Filter, Piecewise liner smoothing, Piecewise constant
smoothing.
131 http://doi.org/10.17993/3ctecno.2020.specialissue4.129-139
3C Tecnología. Glosas de innovación aplicadas a la pyme. ISSN: 2254 – 4143 Edición Especial Special Issue Marzo 2020
1. INTRODUCTION
Discovering an unexplained undersea globe has paying attention in modern era. Clear
descriptions in oceanic surroundings take part a signicant task in discovering as well as
inspecting underneath globe, namely to observe oceanic biodiversity, undersea salvage,
perceiving submerged tube drip, undersea computer visualization applications etc. Here
the submerged imaging nds a large-scale application. Some diculty comes all throughout
underneath descriptions of illumination absorption as well as scattering. Deep-rooted design
of ocean also composes complications in undersea. On the design of the ocean, reection
of the illumination alters. The reected radiance is horizontally polarized along with its
halfway gets inside the water vertically. Vertical polarization has signicant property that
hatches the substance not as great shining and aides on the way to capture deep colors. One
more diculty in underneath descriptions associated towards underneath density as 800
bits impenetrable than air. Hence when beam travels from air into water, it is halfway back
and, in the meantime, partially enters the water. As we go deeper into ocean, dimension
of beam underneath starts reducing. The underneath molecules absorb assertive size of
beam and create problem for capturing imagery. That is the reason; undersea descriptions
are getting darker as depth increases. The color with shorter wavelength travels remoteness
as compared to longer wavelength. This is why undersea descriptions conquered only with
blue color as in Torres-Méndez and Dudek (2005), Chiang and Chen (2012).
There are many challenges that enhance visibility of corrupted descriptions. While
weakening of submerged descriptions outcomes the combination of multiplicative as well
as additive procedures in Schettini and Corchs (2010) conventional improvement system
namely contrasts alteration, histogram equalizer is robustly defective for such assignment.
Former mechanism to review in section II, diculty was attempted with customized
attainment strategy by several imagery in Narasimhan and Nayar (2003), specic module in
He and Seet (2004) or divergence methods in Schechner and Averbuch (2007). Regardless
of their accomplishment, above approaches undergo various problems which degrade the
system performance.
In contrast, this paper proposes new method for enhancing undersea descriptions
using Piecewise linear smoothing. Our approach uses Piecewise linear smoothing gradient
132 http://doi.org/10.17993/3ctecno.2020.specialissue4.129-139
3C Tecnología. Glosas de innovación aplicadas a la pyme. ISSN: 2254 – 4143 Edición Especial Special Issue Marzo 2020
guided lter for getting better picture. Numerous spatial eld schemes use bilateral scheme,
which causes blurring and gradient deformation. Gradient Guided representation performs
ltering process using guidance picture substance. Thus, a boundary preservation method
mainly to improve excellence of underneath descriptions.
This paper prepared as follows. Foremost, in Section 2, we concise the existing schemes.
Section 3 introduces a comparison of conventional and projected lters. Section 4 describes
a new method for enhancing undersea descriptions. In Section 5, we describe tentative
outcomes and at last Section 6 conclude our method.
2. RELATED WORK
Edge-preserving smoothing is the fundamental processing procedure within several low-
level computer visualization applications in Farban, Fattal and Lischinski (2010), Farbman,
Fattal, Lischinski, and Szeliski (2008), Gastal and Oliveira (2011; 2012). Meant for on whole
smooth lters believe the smoothed output imagery are piecewise constant. Generally, the
edge-preserving techniques using conned ltering to keep sharp boundaries. Bilateral lter
is extensively used because of its eortlessness Tomasi and Manduchi (1998). Conversely,
it undergoes unwanted sharpening of edges may show undesired proles around edges.
Guided lter introduced in He, Sun, and Tang (2013) overcome these problems but show
unwanted smoothing edges. Weighted GF in Li et al. (2015) uses gradient-domain constrains
for smoothing the picture elements but in few cases, it cannot preserve the boundaries. The
gradient domain GIF in Kou, Chen, Wen, and Li (2015) incorporates a precise initial-order
boundary-aware restraint to keep up boundaries better in some cases.
These conventional schemes are typically denoted as local model which causes artifact such
as gradient reversals, hence may not ne for few cases. For those schemes, a piecewise linear
form preferred mostly for properly smooth out boundaries. So, no artifacts are present in
improved results. In Liu et al. (2018) piecewise linear method via guided representation
accurately resolve diculty of gradient-reversal except that only some cases illustrate small
smoothing boundaries.
Therefore, we project a P-GGF to properly sharp, smooth all boundaries as well as do
artifacts free enhanced result. Three major goals of projected sections as follows:
133 http://doi.org/10.17993/3ctecno.2020.specialissue4.129-139
3C Tecnología. Glosas de innovación aplicadas a la pyme. ISSN: 2254 – 4143 Edición Especial Special Issue Marzo 2020
1. To take away gradient reversal, we use Gradient Guided Filter (GGF) to give sharper
boundaries.
2. Next, we use piecewise linear smoothing to smooth the boundaries.
3. The projected module uses P-GGF to produce improved output.
Experimental results produced by P-GGF can perfectly remove the problems caused by the
conventional method.
3. PIECEWISE LINEAR SMOOTHING
Conventional methods mostly suitable for image regions more likely be piecewise linear may
cause artifacts. Clearly, detail layers cannot correctly say details in the original descriptions.
Thus, gradient reversal artifacts exist in their enhanced imagery. In highlighted regions,
two kinds of smoothing can properly keep strong gradients which means strong edges
are properly preserved. For weak edges of small gradients which should be smoothed
out. However, these gradients are not properly smoothed by conventional methods or
even improperly sharpened. As a result, gradient reversal artifacts exist with the enhanced
representation.
For most cases, conventional methods usually need a huge number of bins to avoid
quantization artifacts. Smoothed and enhanced descriptions achieved with rst smoothing
of conventional lters and later uses smooth gradient for reconstructing the image. The
reason for these phenomena is clear. The classical smoothing performed in intensity domain
where intensity values could be very large. In contrast, to overcome all problems we go for
proposed method namely piecewise linear method.
Input Image Transformation Decomposition Scalling
Post
processing
Enhanced
Image
Figure 1. Block diagram of Proposed Method.
134 http://doi.org/10.17993/3ctecno.2020.specialissue4.129-139
3C Tecnología. Glosas de innovación aplicadas a la pyme. ISSN: 2254 – 4143 Edición Especial Special Issue Marzo 2020
4. PROPOSED METHOD
As depicted in Figure 1 our technique is mainly for enhancing underwater images using
piecewise linear smoothing. Initially, original picture developed by RGB or HSV color
space. RGB depict colors in combination of primary colors.
HSV describes colors namely Hue, Saturation, and Value. Color depiction acting an
essential task, HSV form often chosen over RGB form and express color likewise to how
human vision lean-to recognize color. Later, we used rst layer decomposition by smoothing
the image using L1L0 smoothing and second layer decomposition for sharpening the edges
using gradient guided lter. Next, scaling to vary ocular form of a picture. Finally, post
processing of scaled image to enhance the quality of the image.
The conventional method can simply modeled as:
(1)
where
represents expected output pixel rate at location
, denotes expected
constant rate of pixel values inside kth area in the picture indicated as
. In contrast to
conventional model assumption, projected method can signify:
(2)
where
denotes expected output pixel value at position p, represents pixel value of
guidance image at position p.
and b
k
stay steady in
. These methods show no
gradient reversal artifacts.
In this paper, we assume a linear form which is spatially linear dierent from gradient
guided image lter. In addition, our method also focuses on how to do projected method
using classical methods.
Dierent from piecewise linear function, we formulate imagery as:
(3)
135 http://doi.org/10.17993/3ctecno.2020.specialissue4.129-139
3C Tecnología. Glosas de innovación aplicadas a la pyme. ISSN: 2254 – 4143 Edición Especial Special Issue Marzo 2020
where as well as b
k
denotes some linear coecients which assumed to stay
constant inside
. The above equation indicates linear polynomial regression to remove
stair-casing eect. The model in (3) is general and abstract form which is not used for
explicit formulations of lters. However, when we take the derivative of
with respect
to p, then we have:
(4)
Note that
signies gradient of picture at p. For linear form, their gradients are piecewise
constant. Though, dissimilarity assumes resultant imagery of smoothed output gradients.
Conversely, problem is that we cannot simply reconstruct picture only by its smoothed
gradients. The diculty in using original and its ltered gradients is to reconstruct ltered
image. For original picture, its x-axes and y-axes gradients denoted as
and . By
denoting smoothing process of piecewise constant lters as
, then nal representation
reconstructed by minimizing the following energy function:
(5)
From this, we can execute piecewise linear smoothing through conventional scheme in the
following two steps:
(1). Smoothing x-axes in addition to y-axes, gradients
and of original representation
with . The smoothed output gradients are denoted by and
(2). Using (5) for enhancing picture
from
,
and
with a proper
value of β. Then the enhanced
is spatially piecewise linear as modelled in (3).
5. EXPERIMENTAL RESULTS
This section presents a comparison of improved outcome produced via ours and conventional
schemes through subjective evaluation of test scene. Figure 2 gives improved result than
conventional methods. As seen from Figure 2(a)-(f), all schemes get better output to some
136 http://doi.org/10.17993/3ctecno.2020.specialissue4.129-139
3C Tecnología. Glosas de innovación aplicadas a la pyme. ISSN: 2254 – 4143 Edición Especial Special Issue Marzo 2020
extent. Figure 2(a) bilateral process Tomasi and Manduchi (1998) results enhanced scene,
but it cannot keep sharp boundaries. In Figure 2(b) He et al. (2013) discussed Guided lter to
keep boundaries sharper but fail to represent the picture well close to a few boundaries. To
defeat Li, Zheng, Zhu, Yao and Wu (2015) used weighted guided scheme in Figure 2(c) used
for minimizing halo artifacts but fail to keep the boundaries.
f) Own elaboraon
Figure 2. Enhancing performance evaluation of Underwater Imagery.
Figure 3. Various applications of Proposed Method a) Smoothing b) Flash c) Feathering.
137 http://doi.org/10.17993/3ctecno.2020.specialissue4.129-139
3C Tecnología. Glosas de innovación aplicadas a la pyme. ISSN: 2254 – 4143 Edición Especial Special Issue Marzo 2020
Figure 2(d), Kou et al. (2015) used an edge aware factor introduced to keep up well-built
boundaries but smoothed away weak boundaries. Piecewise linear method via guided
representation Liu et al. (2018) in Figure 2(e) accurately resolve diculty of gradient-reversal
except that only some cases illustrate small smoothing boundaries. At last projected approach
Figure 2(f) outcomes strong and weak boundaries accurately and makes the standardized
system into range [0; 1] earlier to smoothing and then standardize back towards original
range after smoothing. Figure 3 gives various applications of projected scheme to smooth out
weak boundaries, drop artifacts using ash ltering and rene boundaries using feathering.
However, the projected scheme gives improved results than the schemes in Tomasi and
Manduchi (1998), He et al. (2013), (Li et al., 2015), Kou et al. (2015), Liu et al. (2018).
Therefore, projected scheme is more suitable for various applications.
6. CONCLUSIONS
We projected a piecewise linear smoothing Gradient Guided Filter (P-GGF) method to
improve undersea descriptions. The proposed method properly handles gradient reversal
artifacts caused by conventional and uses GGF with sharper boundaries and smoothed
result based on piecewise linear model. Overall, projected scheme P-GGF can eectively
improve the scene. Experimental results prove that projected method generate imagery
by improved ocular excellence than conventional methods. We believe that proposed
technique used for many applications such as smoothing, Flash and feathering. The extra
remarkable diculty based on the expansion of the projected method to obtain ne details
from numerous descriptions concurrently using extensive lter in Li et al (2012; 2014). We
leave this for future research.
ACKNOWLEDMENT
We gratefully thank the Visvesvaraya Technological University, Jnana Sangama, Belagavi
for nancial support extended to this Research work.
138 http://doi.org/10.17993/3ctecno.2020.specialissue4.129-139
3C Tecnología. Glosas de innovación aplicadas a la pyme. ISSN: 2254 – 4143 Edición Especial Special Issue Marzo 2020
REFERENCES
Chiang, J. Y., & Chen, Y. C. (2012). Underwater image enhancement by wavelength
compensation and dehazing. IEEE Transactions on Image Processing, 31(4), 1756-1769. https://
doi.org/10.1109/TIP.2011.2179666
Farban, Z., Fattal, R., & Lischinski, D. (2010). Diusion maps for edge-aware image
editing. ACM Transa Grap. https://www.cse.huji.ac.il/labs/cglab/projects/dimaps/
dimaps.pdf
Farbman, Z., Fattal, R., Lischinski, D., & Szeliski, R. (2008). Edge-preserving
decompositions for multi-scale tone and detail manipulation. ACM Transactions on Graphics,
27(3). https://doi.org/10.1145/1360612.1360666
Gastal, E. S. L., & Oliveira, M. M. (2011). Domain transforms for edge aware-
image and video processing. ACM Transactions on Graphics, 30(4). https://doi.
org/10.1145/2010324.1964964
Gastal, E. S. L., & Oliveira, M. M. (2012). Adaptive manifolds for real-time high-dimensional
ltering. ACM Transactions on Graphics, 31(4). https://doi.org/10.1145/2185520.2185529
He, D. M., & Seet, G. G. L. (2004). Divergent-beam LiDAR imaging in turbid water. Optics
and Lasers in Engineering, 41(1), 217-231. https://doi.org/10.1016/S0143-8166(02)00138-0
He, K., Sun, J., & Tang, X. (2013). Guided image ltering. IEEE Transactions on Pattern
Analysis and Machine Intelligence, 35(6), 1397-1409. https://doi.org/10.1109/TPAMI.2012.213
Kou, F., Chen, W., Wen, C., & Li, Z. (2015). Gradient Domain Guided Image Filtering.
IEEE Transactions on Image Processing, 24(11). https://doi.org/10.1109/TIP.2015.2468183
Li, Z., Zheng, J., Zhu, Z., Yao, W., & Wu, S. (2015). Weighted guided image ltering. IEEE
Transactions on Image Processing, 24(1), 120-129. https://doi.org/10.1109/TIP.2014.2371234
Li, Z. G., Zhe, J. H., & Rahardja, S. (2012). Detail-enhanced exposure fusion. IEEE
Transactions on Image Processing, 21(11). https://doi.org/10.1109/TIP.2012.2207396
139 http://doi.org/10.17993/3ctecno.2020.specialissue4.129-139
3C Tecnología. Glosas de innovación aplicadas a la pyme. ISSN: 2254 – 4143 Edición Especial Special Issue Marzo 2020
Li, Z.G., Zheng, J., Zhu, Z., & Wu, S. (2014). Selectively detail enhanced fusion of
dierently exposed images with moving objects. IEEE Transactions on Image Processing, 23(10).
https://doi.org/10.1109/TIP.2014.2349432
Liu, W., Xu, W., Chen, X., Huang, X., Shen, C., & Yang, J. (2018). Edge Preserving
Piecewise Linear Smoothing using Piecewise Constant Filters. arXiv:1801.06928. https://
arxiv.org/abs/1801.06928
Narasimhan, S. G., & Nayar, S. K. (2003). Contrast restoration of weather degraded
images. IEEE Transactions on Pattern Analysis and Machine Intelligence, 25(6),713- 724. https://
doi.org/10.1109/TPAMI.2003.1201821
Schechner, Y. Y., & Averbuch, Y. (2007). Regularized image recovery in scattering media.
IEEE Transactions on Pattern Analysis and Machine Intelligence, 29(9), 1655–1660. https://doi.
org/10.1109/TPAMI.2007.1141
Schettini, R., & Corchs, S. (2010). Underwater Image Processing: State of the Art
of Restoration and Image Enhancement Methods. EURASIP Journal on Advances in Signal
Processing, 746052. https://doi.org/10.1155/2010/746052
Tomasi, C., & Manduchi, R. (1998). Bilateral ltering for gray and color images. In
Proceedings of the 1998 IEEE International Conference on Computer Vision, Bombay, India. https://
users.soe.ucsc.edu/~manduchi/Papers/ICCV98.pdf
Torres-Méndez, L. A., & Dudek, G. (2005). Color Correction of Underwater
Images for Aquatic Robot Inspection. In Rangarajan A., Vemuri B., & Yuille A.L.
(eds). Energy Minimization Methods in Computer Vision and Pattern Recognition. EMMCVPR
2005. Lecture Notes in Computer Science, vol. 3757. Springer, Berlin, Heidelberg. https://doi.
org/10.1007/11585978_5
