Journal of Science and Technology

ISSN: 2456-5660 Volume 6, Issue 3, May-June 2021

www.jst.org.in DOI: https://doi.org/10.46243/jst.2021.v6.i3.pp62-66

Enhancement of Underwater Images Using FPGA

J.Sneha Latha1, G.Usharani2
1,2(Department of Electronics and Communication, MGIT, Hyderabad, Telangana)
1Corresponding Author: [email protected]

To Cite this Article

J.Sneha Latha1, G.Usharani2“Enhancement of Underwater Images Using FPGA”, Journal of Science and
Technology, Vol. 06, Issue 03, May-June 2021, pp.:62-66.

Article Info
Received: 12-05-2021 Revised: 15-05-2021 Accepted: 18-05-2021 Published: 22-05-2021
ABSTRACT:
Due to low illumination and changing environmental circumstances, underwater photography is difficult. The obtained weak images must be
upgraded due to limited light offered by the Automated Underwater Vehicles (AUV). In order to employ AUVs, image enhancing algorithms
must be created in hardware. Because of their intrinsic parallelism, Field Programmable Gate Arrays (FPGA) have shown to be a superior
solution for image processing approaches. Power law transform, contrast stretching, Histogram Equalization, Histogram Stretching, and Negative
transformation are among the spatial domain procedures used to preprocess an underwater picture. A comparison of the above-mentioned picture
enhancement technique implemented in FPGA is conducted in this work.

FPGA, Image Enhancement, Histogram, and Power Law Transform are all terms used in this research.

I. Introduction
There is a rich history of undersea culture in the subaquatic environment. Underwater archaeology is the study of this rich cultural past under the
surface of the oceans. Underwater photography has captured coral reefs, oil pipelines, telephone cables, and shipwrecks. [1] As a result, deep-sea
research is currently being carried out. In ancient times, the sea was the primary route of transportation. They have had an enormous impact on
trade. Shipwrecks were widespread in those days because of the ever-changing and unpredictable weather patterns in these areas. As a result,
shipwrecks provide archaeologists with a window into the past. As time capsules, shipwrecks may reveal a lot about the people and culture of the
period. Archaeological practises, even those based on science, may lead to the destruction of valuable artefacts.

Due to the absence of light, underwater photography is challenging. Suspended particles and molecules in water scatter and absorb light rays as
they move through the water's surface. As the depth of the water grows, the quality of the images improves. Artificial lighting is necessary
because of the dispersion of light in the aquatic environment. As a result, the region is illuminated unevenly by this artificial light. Any sensor
that generates electromagnetic radiation with spatially scattered intensity values that can be processed and stored in RAM may create pictures.
[2] Each application requires a different approach for obtaining images, such as how quickly they can be taken and how precise they can be, as
well as a wide variety of other factors.

If you want your images to be as precise as possible, you'll need additional software and patience. FPGAs and ASICs are more beneficial to
designers (Field Programmable Gate Arrays and Application Specific Integrated Circuits). The cost of the tool, as well as its design flexibility
and performance, are important considerations. FPGAs are preferred over ASICs because of their field programmability, reusability, faster
design cycle, and cheaper cost. Performance may be boosted by using FPGAs' specific hardware features like as high-speed I/O and block RAM.

Multilevel contrast stretching for picture enhancement was suggested by the authors of Xu et al [3]. Divides the picture into its individual
components, then improves contrast at both the inter-object and intra-object levels. Artefacts like ringing and blocking may be prevented from
being introduced. Image sequence contrast may be improved using a block-overlapped histogram equalisation technique, as described by Tae et
al. (1998)[4]. Real-time implementations of the classic histogram-based contrast enhancement approach are hampered by the high processing and
storage needs and the possible loss of seldom distributed pixel intensities, which might result in catastrophic data loss. It is clear that Jiang Duan
[5], the protagonist of Jiang Duan, is a fictitious figure. An example of how a local tone mapping approach may be utilised to generate high
dynamic range pictures on a common display is shown in this article. In order to mimic local contrast prior to contrast stretching, it uses bilateral
filtering to smooth out the picture and adaptively stretch the contrast in local locations. Neighbourhood-based equalisation of histograms
According to Eramian et al. [6,7], the histogram equalisation approach suggested by Eramian and his colleagues is unique in that the histogram is
further categorised using local visual features. The sort of measure to use is influenced by how the bins are split and whether or not contrast
enhancement is provided.

A brief introduction to different image processing algorithms is provided in the following sections before a more in-depth look at the hardware
implementation, results, and debate. The histogram equalisation method produces a superior outcome by evaluating the quality measurement
factors.

Preliminary Details
Colorometric and generic picture perspectives are two topics of image processing that have lately received a lot of attention. Each point in the
image has a different intensity based on its location in space and is stored as a numerical value indicating its memory strength. Grayscale pictures
need less storage space than colour photos.

Published by: Longman Publishers www.jst.org.in 62 | 5

 Journal of Science and Technology
ISSN: 2456-5660 Volume 6, Issue 3, May-June 2021

www.jst.org.in DOI: https://doi.org/10.46243/jst.2021.v6.i3.pp62-66

T is used to transform f(x, y) into g by applying it to an image f(x,y) (x, y). As seen in this equation, the values of the two pixels are linked by the
following equation: g(x, y) = Assuming that T is an operator on the pixel values (x,y) defined around point (g(x,y)), this equation connects the
two values (f(x,y)) (x, y). [7]

1.1. Image Inverting: Each byte in a grayscale picture has an intensity value ranging from 0 to 255, with 0 being the brightest shade and
255 being the darkest shade. Simply subtract 255 from the current pixel value to invert the picture. It may be expressed
numerically as:

Where g(x,y) is the transformed pixel. L is the maximum intensity level [7].

2.2 Thresholding: Thresholding is a basic picture segmentation technique. Different threshold values may be set to each RGB plane to threshold
a colour picture, and the threshold image can then be combined.

2.3. Power law transform: Image intensity values may be changed by multiplying current intensity values by nth power and nth root.

This may be stated numerically as:

Although the range of dark input values may be restricted, gamma-fractional power law curves may transform this into a wider range of output
values.

2.4 When it comes to stretching, things are rather different: The histogram's spread is the simplest way to think about contrast. The dynamic
range of a photograph encompasses all of the image's brightness variations. Contrast is weaker when several pixels are in the low-intensity zone.
This is one of the simplest methods for gathering data in low-intensity regions. The only way to boost contrast is to transmit the pixels from the
input to the output in a straight line. A and B are the lower and upper bounds of the expanded intensity values, respectively. Unprocessed
versions of the image are represented by the c and d swatches. It is possible to linearly stretch the input pixels to span the whole range if their
range is constrained. [8] This may be expressed numerically as follows:

2.5 Histogram Equalization: The histogram of a digital image with gray level in the range [0, L-1] is a discrete function of the form:

Published by: Longman Publishers www.jst.org.in 63 | 5

 Journal of Science and Technology
ISSN: 2456-5660 Volume 6, Issue 3, May-June 2021

www.jst.org.in DOI: https://doi.org/10.46243/jst.2021.v6.i3.pp62-66

Probability P(rk) of occurrence of rk grey levels.

In a normalised histogram, the total of all components equals one.

The histogram's components are more concentrated at the bottom end of the grey scale in the dark picture.

It is standard practise to use histogram equalisation to improve the look of photographs. Using equalisation of histograms, an image's intensity
may be altered. At f(x,y), the picture's intensity is at its highest (x,y). From 0 to L-1, the total number of intensity levels accessible, the intensity
levels range from 0 to L-1.

For each pixel with an intensity of n, divide its total number of pixels by the value of Pn. g(x,y) is an equalised histogram picture with the
definition given below.

2.6. Median Filtering: In order to reduce unwanted pixel intensities, pictures must be preprocessed. Pixel by pixel, the median of the pixels
around it is used to replace each individual pixel value in the image. Because there are no beginning and ending pixel values, each of these pixels
must be repeated. As a result, median filtering has a number of important downsides, including these. Median filtering preserves edges while
greatly reducing noise.

This technique, also known as averaging filtering, uses the average intensity of neighbouring pixels to replace each pixel's intensity. F(x,y) is the
intensity of the input picture at (x,y), and g(x,y) is the modified image. The following is a numerical representation of average filtering:

II. Design and Hardware Implementation

Xilinx is the best choice for image processing, according to the study [10]. Since the Altium Designer software and Xilinx Spartan 3 AN are
used, the hardware design is done on the Altium Nano board 3000 as a development platform. Customized hardware is built for the application.

III. Result and Discussion

Underwater imaging study has shown that extra lighting is required due to water molecules absorption of light wavelengths. Blue and green
images are created when light waves penetrate deeper in the open ocean, while green images are created when light waves penetrate deeper in
coastal seas. The colour qualities of photos are affected by the extra light given by imaging equipment. Colorful photos make up most of the
collection. As a result, one of the RGB channels in the picture has become more saturated. For each pixel in this study, each colour component is
extracted, and the process is repeated for each one. It is possible to do per-pixel processes using this approach, which limits the gain of saturated

Published by: Longman Publishers www.jst.org.in 64 | 5

 Journal of Science and Technology
ISSN: 2456-5660 Volume 6, Issue 3, May-June 2021

www.jst.org.in DOI: https://doi.org/10.46243/jst.2021.v6.i3.pp62-66

pixels. The FPGA has also been used to develop image processing methods like inversion and thresholding. Numerous methods for improving
performance were tested. Various underwater photograph-enhancing approaches are examined in Tables 1 and 2 based on their Peak Signal to
Noise Ratio and Structure Similarity Index Matrix (SSIM). The Histogram equalisation technique is outperformed by other alternatives in terms
of performance.

Published by: Longman Publishers www.jst.org.in 65 | 5

 Journal of Science and Technology
ISSN: 2456-5660 Volume 6, Issue 3, May-June 2021

www.jst.org.in DOI: https://doi.org/10.46243/jst.2021.v6.i3.pp62-66

IV. Conclusion
One of the colour channels in underwater photos is more saturated than the other two. Because of the non-uniform illumination and light
propagation qualities, the green and blue channels are saturated in the tested photos. The algorithm's resilience has been evaluated on a large data
set to ensure that it is reliable. There are no unappealing artefacts when using the power law conversion with varied gamma values. FPGAs have
a huge number of memory blocks that may be accessed in parallel. As a result, the time it takes to process each pixel is significantly reduced. The
FPGA is a superior option for image processing applications because of its cheap cost and high performance. The goal of this project was to
successfully incorporate fundamental image processing algorithms for underwater photographs on the Altium Nano board 3000. As a
consequence of the foregoing results on evaluated photos, the FPGA is required in the area of image processing.
References
[1]. Retrieved from : http://www.bbc.co.uk/history/ancient/archaeology/marine_01.shtml
[2]. Erhardt – Ferron ;Theory and Applications of Digital Image Processing university Of Applied sciences of fenburg hochschule for technik
undwirt schaft
[3]. Beilei Xu; Yiqi Zhuang; Hualian Tang; Li Zhang, "Object-based multilevel contrast stretching method for image enhancement," Consumer
Electronics, IEEE Transactions on , vol.56, no.3, pp.1746,1754, Aug. 2010.
[4]. Kim, Tae Keun, Joon Ki Paik. and Bong Soon Kang. Contrast enhancement system using spatially adaptive histogram equalization with
temporal filtering. Consurner Electronics, IEEE Transactions on 44, no. I (1998): 82-87
[5]. Jiang Duan; Wenpeng Dong; Rui Mu; Guoping Qiu; Min Chen, "Local contrast stretch based tone mapping for high dynamic range images,"
Computational Intelligence for Multimedia, Signal and Vision Processing (CIMSIVP), 2011 IEEE Symposium on , vol., no., pp.26,32, 11-15
April 2011
[6]. Eramian, M.; Mould, D., "Histogram equalization using neighborhood metrics," Computer and Robot Vision, 2005. Proceedings.
The 2nd Canadian Conference on , vol., no., pp.397,404, 9-11 May 2005
[7]. Rafael C. Gonzalez, Richard E. Woods, ``Digital Image Processing,'' 2nd Edition, Prentice Hall; ISBN: 0201180758; January 15, 2002.
[8]. Jagatheeswari, P.; Kumar, S.S.; Rajaram, M., "Contrast Stretching Recursively Separated Histogram Equalization for Brightness Preservation
and Contrast Enhancement,"Advances in Computing, Control, & Telecommunication Technologies, 2009. ACT '09.
International Conference on ,vol.,no.,pp.111,115,28-29Dec.2009doi: 10.1109/ACT.2009.37
[9]. K.S. Sim, C.P. Tso, Y.Y. Tan, Recursive sub-image histogram equalization applied to gray scale images, Pattern Recogn. Lett. 28 (2007)
1209–1221.
[10]. Shrikanth G., and Kaushik Subramanian. "Implementation Of Fpga•Based Object Tracking Algorithm." Postgraduate Thesis,(SVCE, India)
(2008).Vol. 1. IEEE, 2009

Published by: Longman Publishers www.jst.org.in 66 | 5