2024 2nd World Conference on Communication & Computing (WCONF)

Raipur, India. Jul 12-14, 2024

Implementation of Sobel Edge Detection with

Image Processing on FPGA
Dr. V. Priyanka1 Y. Sri Rama2 Kadaru Sravani3
Asst. Prof., Dept. of ECE Asst. Prof., Dept. of ECE, Dept. of ECE,
Vallurupalli Nageswara Rao Vignana Vallurupalli Nageswara Rao Vignana Vallurupalli Nageswara Rao Vignana
Jyothi Institute of Engineering and Jyothi Institute of Engineering and Jyothi Institute of Engineering and
Technology, Technology, Technology,
Hyderabad, India Hyderabad, India Hyderabad, India
priyanka_v@vnrvjiet.in srirama_y@vnrvjiet.in sravanikadaru.7@gmail.com

Basam Kavya4
Dept. of ECE,
Vallurupalli Nageswara Rao Vignana
2024 2nd World Conference on Communication & Computing (WCONF) | 979-8-3503-9532-7/24/$31.00 ©2024 IEEE | DOI: 10.1109/WCONF61366.2024.10692301

Jyothi Institute of Engineering and

Technology,
Hyderabad, India
kavya.basam@gmail.com

Abstract— From laptop vision to medical diagnostics, throughput quotes. Thus, Sobel part detection emerges as a
picture processing is essential in lots of fields. In order to cornerstone technique in the realm of photograph processing,
improve actual-time overall performance, this take a look at supplying a sturdy mechanism for extracting vital visual cues
makes a speciality of applying the Sobel side detection technique from virtual imagery.
on a Field-Programmable Gate Array (FPGA) platform and
taking advantage of its parallel processing abilities. One method II. RELATED WORK
this is regularly used to discover edges and lines in pics is known
as Sobel side detection. Our recommended FPGA-based
Reference [1], This paper proposes a novel edge detection
technique includes using a Hardware Description Language method combining Sobel and median filtering. Experimental
(HDL), which include Verilog, to layout and enforce a selected consequences demonstrate stepped forward accuracy and
hardware structure. The layout uses Sobel convolution kernels robustness through reducing noise and retaining area
to calculate gradient values and aspect importance through information. This fusion technique offers capacity
applying the kernels to photograph pixels in each horizontal and advancements in various fields reliant on particular part
vertical orientations. Optimizing parallel processing for detection in virtual imagery. Reference [2], This paper delves
effective real-time performance is one of the maximum vital into the efficacy of Sobel edge detection on photo edges
factors. After everything is completed, the display and through MATLAB experimentation. Authors systematically
photograph acquisition modules are included into the device. compare the Sobel operator's performance throughout
Performance evaluation includes measures like throughput, numerous eventualities, scrutinizing edge sharpness, noise
latency, and aid utilization to assure the effectiveness of the sensitivity, and computational performance. Insights garnered
Sobel aspect detection machine primarily based on FPGA. shed light on Sobel's strengths and obstacles in edge detection.
Reference [3], This paper introduces a pioneering approach to
Keywords— Sobel edge, FPGA, edge detection, gradient edge detection utilizing the Sobel filter deployed on an FPGA
values platform. Experimental validation showcases the viability and
I. INTRODUCTION efficacy of FPGA-driven edge detection, leveraging parallel
processing for real-time operation. Sobel's hardware
Within the world of digital image processing, the venture implementation on FPGA resources demonstrates reduced
of edge detection holds paramount significance, serving as an computational latency, enhancing efficiency compared to
essential precursor to a myriad of laptop vision endeavors. software-based counterparts. Reference [4], This study
Among the multitude of strategies available, the Sobel comprehensively examines Sobel edge detection,
operator emerges as a distinguished preference, respected for encompassing its algorithmic foundations, implementation
its simplicity and computational performance. Comprising methodologies, and performance evaluation metrics. Through
two 3x3 convolution kernels—one tailor-made for detecting a blend of theoretical analysis and practical experimentation,
horizontal modifications and the other for vertical changes— the study scrutinizes Sobel's effectiveness in delineating edges
the Sobel operator intricately calculates gradient and contours in digital images. Considerations such as
approximations along both the x and y axes. Through the computational complexity, edge detection accuracy, and noise
amalgamation of these gradients, it furnishes a gradient robustness are discussed, elucidating the Sobel operator's
importance image, adept at accentuating the edges embedded strengths and limitations. Reference [5], This paper delineates
inside the original visual facts. Noteworthy is the Sobel the hardware architecture tailored for FPGA implementation
operator's inherent simplicity and computational efficacy, of the Sobel edge detection algorithm, elucidating data flow,
attributes that render it particularly well-appropriate for real- processing elements, and memory requisites. Implementation
time programs and integration inside embedded structures. specifics, encompassing HDL code, clock frequency, and
Moreover, its linear nature lends itself seamlessly to resource utilization, are meticulously outlined. Additionally,
implementation inside hardware architectures, appreciably on considerations for optimizing performance and resource usage
Field-Programmable Gate Arrays (FPGAs), where parallel
processing skills can be harnessed to reap commendable

979-8-3503-9532-7/24/$31.00 ©2024 IEEE 1

Authorized licensed use limited to: VIT University. Downloaded on November 08,2024 at 14:51:13 UTC from IEEE Xplore. Restrictions apply.
are discussed, offering insights into efficient FPGA-based 3. Robert Operator: This gradient based operator primarily
Sobel edge detection. makes use of discrete differentiation to calculate the sum of
the squares of the variations between the diagonally adjoining
III. OBJECTIVES pixels within the photo. After that, the gradient is
The goal of this undertaking is to develop and put into approximated. The following 2 × 2 kernels or mask are being
effect a real-time Sobel side detection machine on FPGA for used.
embedded imaginative and prescient systems. This device 1 0 0 1
objective is to accurately perceive edges in input photos at the ‫ = ݔܯ‬ቂ ቃ ‫ = ݕܯ‬ቂ ቃ
0 െ1 െ1 0
same time as meeting stringent performance necessities,
including high overall performance, low latency, and minimal
useful resource usage. Key targets include optimizing the D. Gaussian based types
FPGA architecture and set of rules implementation for 1. Marr-Hildreth Operator or Laplacian of Gaussian
efficiency, integrating the gadget with appropriate input and (LoG): It involves deriving the second spinoff of an image
output interfaces for seamless integration into embedded using Gaussian-based totally operator. It plays successfully
vision structures, and ensuring robustness and reliability via whilst there are surprising transitions inside the grayscale
thorough trying out and verification. Ultimately, the purpose areas of the photograph. This technique operates by using
is to create a versatile FPGA-based totally side detection detecting 0 crossings, wherein the second-order derivative
answer able to deployment throughout loads of simulation crosses 0, indicating the most efficient facet role. By
applications in embedded imaginative and prescient combining the Laplacian operator to lessen noise and the
structures. Gaussian operator to identify sharp edges, it enhances side
A. Edge Detection detection accuracy.
In the area of picture processing, area detection performs A formula defines the Gaussian function:
an essential position in delineating the limits of a photograph.
These edges represent exquisite shifts in depth or color, 1 ‫ ݔ‬ଶ + ‫ݕ‬ଶ
‫ݔ(ܩ‬, ‫= )ݕ‬ ݁‫ ݌ݔ‬െ ቆെ ቇ
generally aligning with the outlines of objects and enormous ξ2ߨߪ ଶ 2ߪ ଶ
functions within the photograph. Detecting these transitions is
imperative for duties like segmentation, item recognition, and ߲ଶ ߲ଶ ‫ ݔ‬ଶ + ‫ ݕ‬ଶ െ 2ߪ ଶ ‫ ݔ‬ଶ + ‫ݕ‬ଶ
complete photograph analysis, as it aids in identifying critical ‫= ܩ݋ܮ‬
߲‫ݔ‬ ଶ
‫ݔ(ܩ‬, ‫ )ݕ‬+ ଶ ‫ݔ(ܩ‬, ‫= )ݕ‬
߲‫ݕ‬ ߪ ସ
exp ቆെ
2ߪ ଶ
ቇ
gaps in the visible information.
B. Edge detection types: 2. Canny Operator: It's a gaussian based operator which
x Gradient based operator: This sort of operator could come across edges. Noise isn't always a hazard to this
computes first-order derivations in a digital image. operator. It extracts the image features and does no longer
have an effect on or modify them. Canny side detector has
It is similarly labeled as Sobel operator, Prewitt superior algorithm derived from the previous paintings of
operator, Robert operator. Laplacian of Gaussian operator. It is extensively used as the
most appropriate aspect detection method. Based on three
x Gaussian based operator: This type of operator standards, it'll be capable of hit upon the rims:
computes 2nd-order derivations in a virtual
photograph. 1.Low error fee
It is similarly classified as Canny edge detector, 2.Edge points should be correctly localized
Laplacian of Gaussian.
3.There should be just one single edge reaction
C. Gradient based types
E. Sobel Edge Detection
1. Sobel Operator: This operator for differentiation is
The Sobel operator may be just like Prewitt's. It's
wonderful. It computes a gradient approximation of the image
additionally a derivate mask, so it is able to be applied for part
intensity feature for picture edge detection. The Sobel operator
detection. In the same way of the Prewitt operator, Sobel
generates the corresponding gradient vector or the regular to a
operator is used to locate two kinds of edges within the
vector at the pixels of a photo. Two 3 x 3 kernels, or mask,
picture:
made out of an input image, are used to calculate the vertical
and horizontal derivative approximations. x Vertical direction
െ1 0 1 െ1 െ2 െ1 x Horizontal direction
‫ = ݔܯ‬൥െ2 0 2൩ ‫ = ݃ܯ‬൥ 0 0 0൩
The horizontal edges within the image are highlighted
െ1 0 1 1 2 1 while the masks are implemented. It's just as if it has been first
order derivate, calculating the difference in pixel intensities at
2. Prewitt Operator: This operator is sort of precisely an aspect region. Since the middle columns are zero, they do
similar to the Sobel operator. It also figures out the image's not include original values of a picture but merely calculate a
horizontal and vertical edges. It's some of the great strategies difference between proper and left pixel values on that facet.
for figuring out the importance and orientation of images. It The center values are 2 and -2 for both the columns 1 and 3.
uses masks or a kernel. As a result, the pixel values round the threshold region are
െ1 0 െ1 െ1 െ1 െ1 heavier. This will increase the brink's depth and, in contrast to
‫ = ݔܯ‬൥െ1 0 െ1൩ ‫ = ݃ܯ‬൥ 0 0 0൩ the original image, makes it extra extreme. Two masks are
used for the detection of the Sobel area, one for the horizontal
െ1 0 െ1 1 1 1

2
Authorized licensed use limited to: VIT University. Downloaded on November 08,2024 at 14:51:13 UTC from IEEE Xplore. Restrictions apply.
detection of the photograph derivatives and the alternative for G. Implementation of the algorithm with FPGA
the vertical detection of the photograph derivatives. There's The Sobel edge detection algorithm is a popular technique
simply one mask and the opposite is circled ninety ranges. The in image edge processing. It is difficult to adjust the
Sobel kernels can also be thought of as 3 × 3 approximations conventional Sobel edge detection method, which is based on
to first-spinoff-of-Gaussian kernels. The implementation of the PC implementation, to the complex image of today. It
the Edge detector consists of RGB to grayscale color area operates at a slower pace and with less gradients. FPGAs are
conversion and line buffers to synchronize H-sync, V-sync and highly reliable and have great parallel processing capabilities
Data-allow indicators. In order to keep an appropriate rows due to their widespread application and integration. Using
and area detection blocks, a clear out and buffer block consists FPGA technology, image edge detection is achievable.
of line buffers. There are two convolution blocks in the facet Several distinct technologies and approaches have been
detector block that perform the operation distinctive by way employed in the development of the Sobel operator-based
of the kernel. In order to gain vertical and horizontal edges, edge detection system to achieve real-time performance.
each row of every kernel is extended by way of the behind These technologies range from specialized CPUs to
schedule elements saved within the line buffers. They'll be programmable logic devices (FPGAs) to parallel
added and the gradient value will be decided. For the reason architectures. Real-time performance is difficult to attain with
of acquiring accurate edges, the gradient magnitude shall be digital signal processors or general-purpose processors, while
reduced by means of a manually adjusted threshold. FPGAs reduce the substantial design labor needed for
F. Sobel Edge Detection Algorithm application-specific integrated circuits. The improved Sobel
edge detection algorithm based on FPGA is implemented,
The traditional Sobel area detection is to do 3 × 3 pixel making the detection results more rapid and accurate. The
area via the corresponding example convolution operation, it’s design is based on the Xilinx software to write the RAM-based
completed that the center pixel gradient values, then as
shift register, multipath parallel adder module, and floating
compared with predetermined threshold gradient values, if is square root module. Through shift and addition to complete
larger than the edge criterion for decisive the pixel photo the multiplication calculation, the gradient computing
aspect pixels, the opportunity isn’t. The standard Sobel template is added.
operator’s convolution example has two, one is detection level
aspect, one is detection vertical part. As shown in Fig. 2, Sobel
area detection set of rules is based totally on gradient edge
detection approach, which reveals edges using horizontal
mask (HM) and vertical mask (VM).
One mask is the transpose of alternative. Let [P], a 3 × 3
sub-window of the photograph is convoluted with the
horizontal and vertical mask to offer horizontal gradient Gx
and vertical gradient Gy as proven within the Fig. 2. These
kernels are considered to retort maximally to edges running
vertically and horizontally relative to the pixel grid. Gx kernel
is used for vertical orientation and Gy for horizontal
orientations and they are perpendicular to each other. Separate
Fig. 2. Hardware architecture of sobel edge detection
gradient components for the photo are acquired for both the
orientations. These gradient components are acquired by Line Buffers, are temporary storage units that hold the
means of applying the kernels for vertical orientation and image data. They are used for synchronizing the image data
horizontal orientation discretely to the input image. The with processing units. So the timing signals within the line
significant cost of the absolute gradient is located by buffers correspond to the horizontal, vertical image data.
combining the gradience of both orientations. The absolute These line buffers which stores multiple rows of image data
gradient importance is: allows the convolution blocks to access pixels beyond the
|G| = ඥ‫ݔܩ‬2 + ‫ݕܩ‬2 (1) current row which are necessary for applying the Sobel masks.
The line buffers process data related to one row at a time and
ܵ‫ܽ( = ݔ‬2 + ܿܽ3 + ܽ4) െ (ܽ0 + ܿܽ7 + ܽ6) (2) while accessing the nest row, the data related to the previous
row is cleared. Convolution block, is the major part of the
ܵ‫ܽ( = ݕ‬0 + ܿܽ1 + ܽ2) െ (ܽ6 + ܿܽ5 + ܽ4) (3) whole application where the actual Sobel edge detection takes
With constant C = 2, this can be approximated as given in place. Each convolution block likely has weights
(4) to compute the absolute magnitude faster |G| = |Gx| + |Gy| corresponding to the Sobel mask (E.g.: -2,0,2 for horizontal
edges). Now these weights are multiplied with the
corresponding pixel values in the image data. Image data,
which is fetched from the line buffers. The products of each
pixel within the mask are summed. Now this sum value
provides the final edge strength value for the pixel. The square
root block is used to find out the absolute gradient value. It
takes the horizontal and vertical gradient values and finds out
the absolute gradient. After finding the absolute gradient of
one Sobel mask parallel adder and shifting operations are used
to apply the Sobel mask to the next set of the pixel values.
Fig. 1. Edge Detection (Gx) Horizontal Gradient (Gy) vertical Gradient Now all the absolute gradient values obtained from the
convolution operations are sent to the threshold where these

3
Authorized licensed use limited to: VIT University. Downloaded on November 08,2024 at 14:51:13 UTC from IEEE Xplore. Restrictions apply.
gradient values are compared with the threshold value unorder
to detect it as an edge.
H. Block Diagram

Fig. 6. RTL view of proposed Sobel edge detection architecture

The PSNR value stands for Peak-Signal-to-Noise-Ratio,

and the SNR value stands for Signal-to-Noise Power-Ratio. A
good PSNR value will be over 40dB. Likewise, a higher SNR
value indicates the level of image quality. The simulation
result in figure 5.1 shows that, the PSNR value increased from
44.0994 to 50.7689. In the same way, the simulation result in
figure 5.2 shows that the PSNR value increased from 44.0829
Fig. 3. The proposed hardware implementation of sobel edge detection
algorithm to 53.3653. This shows a positive result that the image quality
has increased when the PSNR values are considered.

IV. RESULTS
The image tools used is MCR application (MATLAB
application) which converts the GREY scale image into Edge
detected image. The input for the application is a GREY scale
image which will convert the image data into Hexadecimal
text file (input. hex). The hexadecimal file is given as input to
the Sobel edge algorithm and is simulated through Modalism Fig. 7. Simulation Output graph
tool to get the output. The output is a hexadecimal text file
(output. hex), and this file is now given as input to the MCR
application again to get the edge detected output image.

Fig. 8. Synthesis report in Vivado (for Zynq zybo z-10 board)

V. DISCUSSION
Several mathematical procedures are used for digital
images during edge detection in order to identify discrete
locations with an abrupt shift in pixels that signal
Fig. 4. GUI output-1 of the edge detected image
discontinuities or brightness variations of the image. An edge
is a collection of curved lines created from image pixels where
there are significant fluctuations in brightness. A basic method
for image processing that could be used in computer vision
and machine vision applications to separate features and
extract information is edge detection. One method for
identifying discontinuities in grayscale photos could be edge
detection. In picture edge processing, the Sobel edge detection
algorithm is a widely used method. With a slower processing
speed and a lower gradient degree, a traditional Sobel edge
detection approach based on PC implementation has been
utilized, making it challenging to adapt to the more complex
image. FPGAs have a high degree of stability and great
concurrent processing capabilities because of their wide
application and integration. With Field Programmable Gate
Array technology, the edge of an image may be detected. The
Sobel operator based edge detection system has been
Fig. 5. GUI output-2 of the edge detected image developed using several distinct technologies and
methodologies to attain real-time performance. Specialized

4
Authorized licensed use limited to: VIT University. Downloaded on November 08,2024 at 14:51:13 UTC from IEEE Xplore. Restrictions apply.
CPUs, FPGAs, programmable logic devices, and parallel parallelism to maximize FPGA resource usage. There is
architectures are some examples of these technologies. Real- always adeveloping demand for real-time video processing in
time performance is achieved with FPGAs, which is not applications which include surveillance, robotics and
possible with general-purpose processors or digital signal independent systems.
processors. However, application-specific integrated circuits
require more involved design work. An enhanced Sobel edge REFERENCES
detection technique based on FPGAs has been created to [1] Y. Kashyap, A. Vyas, R. Raghuwanshi, R. Sharma, “Edge detection
improve the detection speed and accuracy. The floating square using sobel method with median filter”, Volume 02, Issue 05, May
root module, multipath parallel adders, and RAM-based shift 2015.
register will all be written using Xilinx software. To finish the [2] E.J. Mohammad, M. JawadKadhim, W.I. Hamad, S.Y. Helyel, Study
sobel edge detection effect on the imageedges using matlab, Int. J.
multiplication computation, add the gradient computing Innov. Res. Sci. Eng. Technol. 3 (3) (March 2014).
template by shifting and adding. [3] A.M. Khidhir Dr., N.Y. Abdullah, “FPGA based edge detection using
sobel filte”, Vol. 2: Issue. 1, June-July 2013 pp. 22–32.
VI. CONCLUSIONS
[4] J.M. Elham, J.M. Ahmed, J.M. Zainab, et al., Design study sobel edge
The FPGA target device uses the Xilinx platform to detection, Int. J. Appl. Innov. Eng. Manage. 2 (12) (December 2013)
implement the Sobel edge detection method. There are several 248–253 ISSN 2319–4847.
benefits of implementing Sobel edge detection with image [5] S.D. Brown, R.J. Francis, J. Rose, Z.G. Vranesic, Field-Programmable
processing on FPGA, including reduced power consumption, Gate Arrays, 180, Springer Science & Business Media, 2012.
hardware acceleration, customization, and real-time [6] G Anusha, T. JayaChandra Prasad Dr, D. SatyaNarayana Dr,
Implementation of sobel edge detection on FPGA, Int. J. Comput.
processing. Because of its parallel processing capabilities, Trends Technol. 3 (3) (2012).
FPGA can process images in real time, which makes it ideal [7] R. Muthukrishnan, M Radha, Edge detection techniques for image
for applications where latency is critical, such autonomous segmentation”, Int. J. Comput. Sci. Inf. Technol. (IJCSIT) 3 (6) (Dec
vehicles or robotics. There is room for further optimization 2011).
and enhancement where the researchers can work on the [8] T.A. Abbasi, et al., A proposed FPGA based architecture for sobel edge
quality tuning the set of rules and exploring techniques to detection operator, J. Act. Passive Electron. Devices 2 (2007).
improve its accuracy and performance. This includes [9] R.C. Gonzalez, R.E. Woods, S.L. Eddins, Digital image processing
exploring advanced filtering strategies and making use of using matlab (2004).

5
Authorized licensed use limited to: VIT University. Downloaded on November 08,2024 at 14:51:13 UTC from IEEE Xplore. Restrictions apply.