Content uploaded by F M Javed Mehedi Shamrat
Author content
All content in this area was uploaded by F M Javed Mehedi Shamrat on Mar 29, 2021
Content may be subject to copyright.
Human Face Recognition Using EigenFace, SURF
Methods
*F.M. Javed Mehedi Shamrat1, Pronab Ghosh 2, Zarrin Tasnim3, Aliza Ahmed Khan4,
Md. Shihab Uddin5, Tahmid Rashik Chowdhury6
1,3Dept. of Software Engineering, Daffodil International University
2, 4,5Dept. of Computer Science and Engineering, Daffodil International University
6Dept. of Computer Science and Engineering, Islamic University of Technology
1javedmehedicom@gmail.com
2pronab1712@gmail.com
3Zarrint25@gmail.com
4alizaahmedkhancse22ju@gmail.com
5shihab651951@gmail.com
6eclipse15051@gmail.com
Abstract. One such complicated and exciting problem in computer vision and
pattern recognition is identification using face biometrics. One such application
of biometrics, used in video inspection, biometric authentication, surveillance,
and so on, is facial recognition. Many techniques for detecting facial biometrics
have been studied in the past three years. However, considerations such as
shifting lighting, landscape, the nose being farther from the camera, the
background being farther from the camera creating blurring, and noise present
render previous approaches bad. To solve these problems, numerous works with
sufficient clarification on this research subject have been introduced in this paper.
This paper analyzes the multiple methods researchers use in their numerous
researches to solve different types of problems faced during facial recognition. A
new technique is implemented to investigate the feature space to the abstract
component subset. Principle Component Analysis (PCA) is used to analyze the
features and use Speed up Robust Features (SURF) technique Eigenfaces,
identification, and matching is done respectively. Thus, we get improved
accuracy and almost similar recognition rate from the acquired research results
based on the facial image dataset, which has been taken from the ORL database.
Keywords: Human face detection, Face recognition, Facial detection, Eigen-
face, Methods, SURF.
1 Introduction
One such optical pattern recognition problem is face recognition [1]. After entering a
random image as input in the face recognition system, it will explore the database and
identify the person as output. Usually, a face identification system contains four
components [2] as shown in Fig. 1: detection, alignment, feature extraction, and
International Conference on Pervasive Computing and Social Networking [ICPCSN 2021]
Salem, Tamil Nadu, India, 19-20, March 2021
Pre-Print
2
matching, the refining steps are localization and normalization (face detection and
alignment) before face identification (extraction and matching) is done [3]. Facial
image identification separates the facial region from the background. However, face
tracking equipment [4] is used to trace the recognized faces in a video. Face alignment
focuses on acquiring more precise localization and thus normalizing faces.
On the other hand, face recognition gives a rough measure of every identified face's
position and scale. Facial elements such as eyes, mouth, nose, and facial layout are
situated [5]. Using the location spots, the face image given as input is standardized, in
respect of geometrical characteristics such as pose and size, using geometrical
distortion or transforms. The face goes through more normalization procedures
regarding photometrical characteristics such as illumination and grayscale. After
standardizing the face geometrically and photometrically, attribute selection is
executed to give the necessary information that helps differentiate between faces of
various persons and steady in respect of the geometrical and photometrical variations
[6]. In the case of face matching, the input face's obtained attribute vector is compared
with that of the registered faces existing in the database. Whether a match is found with
sufficient confidence, it offers the name of the face as output and if no match is found,
an anonymous face is specified. Artificial Intelligence has been properly used to
overcome signal processing issues for 20 years [7]. Researchers suggested various
models related to artificial neural networks. It is a tough procedure to figure out the
most reliable neural network model for solving real-life problems.
Fig. 1. The infrastructure of a Face Recognition System.
The following building blocks typically comprise facial recognition systems:
• Face detection. The first and important stage for facial recognition is Face
Detection, which is used to recognize faces in photographs. It is a result of the
identification of artifacts. A face detector locates the whereabouts of the faces
inside an image and if any face is traced then the coordinates of a bounding
box for all of them are returned. This is depicted in Fig. 2(a).
• Face alignment. The objective of face alignment is to scale and similarly trim
facial images using a group of related points situated at certain positions in the
image. Using a landmark detector, a group of facial markers is located in this
method. We intend to look for the optimal affine evolution adjusted in the
reference points for 2D alignment. Two facial images are oriented in Fig. 3(b)
and 3 (c) using the same related points. Face fractalization (varying the posture
of a face to frontal) can be implemented by other critical 3D orientation
algorithms i.e. [16].
3
• Feature Extraction: While portraying the face, the pixel values of a facial
image are mutated into a close-packed and distinguishing attribute vector,
called template. Logically, each face of the same individual should point to
related attribute vectors.
• Face matching: In the face matching process, a similarity score is obtained by
contrasting two templates which show the probability that they are part of the
same individual. Face representation is indeed the most vital element of the
face identification system and the literature review is focused on in Section II.
Fig. 2. (a) Face Detector; (b) Aligned Faces and Reference Points.
The primary features of the current study are:
• In this system, two key features of face recognition such as Eigenface and
SURF have been demonstrated with the help of PCA components.
• Four different types of eigenvectors (6, 10, 20, and 190) have been computed
based on the Euclidean and Manhattan distances.
• Besides, Euclidean and Manhattan distances were also shown the predicted ac-
curacy of five different persons based on various types of input images to make
this unique approach.
• Both SURF (64 and 128) and SIFT (128) are examined on different dimensions
along with the doubled image sizes of those dimensions.
The following sections of this paper contain: comprehensive related works of face
recognition have been added in Section 2. In section 3, a sufficient description of the
collected dataset is given and a descriptive analysis of the introduced features has been
explained with required working diagrams. An intuitive comparison of results is
generated for showing their performance on the given dataset to make it more
understandable with the aid of graphs in section 4. In the final section 5, this paper's
overall idea has been made to show its capability.
2 Relevant Works
Multiple prevailing face recognition types of research use PCA (Eigenfaces) for face
identification. Some of the current works are illustrated. In [8], the authors brought in
4
updated procedures, or scores [9 – 10], for uniformity of the face to make the investi-
gation easier. Scores are calculated using only the pixel data of the images in the data-
base (and the weighted mean of the pixel data). A 3D face database is used to eliminate
undesirable errors in the calculation of uniformity from issues in 2D images, i.e. illu-
mination. Based on the scores, statistical tests [11] are carried out in different subgroups
of the database to differentiate the uniformity of the face, and then the result of face
recognition is compared with similar subgroups. A significant variation in face uni-
formity scores between the subgroups of men and women is observed and the result of
face identification is contrasted. The database is then split into most uniform subjects
and least uniform subjects based on the uniformity scores and the face identification
outcome is contrasted. They realized that using uniformity in face identification, using
the mean-half-face, is helpful for their analysis. They discovered analytical importance
between male and female subjects' face uniformity in the 3-dimensional database, in-
cluding variation in face identification precision [12]. In full face, the least uniform
subjects generate greater face identification precision than the most uniform subjects.
Nevertheless, face identification precision is globally increased when mean-half-face
is used in the experiments instead of full face. A computerized face identification sys-
tem has been made in [13], to analyze the possible use for office door access control.
Eigenfaces' procedure depending on the principal component analysis (PCA) and arti-
ficial neural networks [14] have been used. Training images can be acquired offline
either by pre-recorded and trimmed facial images or online by using the system's face
recognition and identification training components on the actual front-facing images.
For the rotational angle of the person's head from -20 and +20 degrees, the device may
distinguish the face at a reasonable pace at a distance of 40 cm and 60 cm from the
frame. The experimental result confirmed the impact of illumination and stance on the
facial recognition device. The authors used principal component analysis (PCA) in [15],
to obtain attributes of facial images and to implement face identification, sparse repre-
sentation-based classification (SRC) algorithm is employed. Experimental outcome de-
picts that when the ideal illustration is properly scattered, it can be effectively resolved
using convex optimization, which is referred to as an l1-minimization problem.
Furthermore, the homotopic algorithm can efficiently resolve the l1-minimization prob-
lem, hence for figuring out the object classes, sparse coefficients are employed. The
authors suggest a strategy in [16] depending on an information theory method, where
facial images are fragmented into a tiny set of distinctive attribute images known as
“Eigenfaces”, which in reality are the main elements of the preliminary training set of
facial images. Identification is carried out by creating a new image into the subspace
covered by Eigenfaces (“face space”) and then by contrasting the location of the face
in the face space with the location of the face of the known persons, the face is identi-
fied. An effective method to discover the lower-dimensional space is the Eigenface
method. In reality, Eigenfaces are proprietary vectors that represent each dimension in
face space and can be used as different face attributes. Both face forms may be de-
scribed in the face collection as a linear fusion of singular vectors. To be exact, these
singular vectors are the covariance matrices vectors. In displaying the significant char-
acteristics, the eigenfaces played an essential function, thereby reducing the input size
of the neural network.
5
3 Research Methodology
3.1 Data Collection
The test was carried out by using the ORL database (face data) [17]. Within the training
database, there are 190 images of 38 people (5 images for each person) and 40 images
of different people (38 familiar and 2 unfamiliars) are present in the test database. In a
straight-up, front-view posture, a photograph of the subject is taken. A picture has a
similar unilluminated backdrop and 92×112 measurements. Besides, each picture is
grayscale (intensity measures of gray are considered image attributes).
3.2 PCA Approach to Face Recognition
A sequence of data derived from logically linked variables [18] is converted by key
component analysis into a collection of values for non-correlated variables called main
components. The number of components may be smaller than or equal to the initial
number of variables [19]. The first major variable has the largest potential variance.
Under the constraint that it has to be orthogonal to the previous component, each of the
effective components has the greatest possible variance [20]. We want to find the key
components of the covariance matrix of facial images, in this case, eigenvectors. The
first thing we need to do is to build a data set for training. The fundamental techniques
followed by this have been depicted in Fig. 3 to show the overall working process.
Fig. 3. Flow Diagram of PCA Approach
At first, the collected images were separated into two parts: training and testing to cal-
culate the covariant value. After that, both eigenvector and eigenvalues were computed
before getting the results from the covariant matrix. In the following stage, when the
6
matrix was calculated, the face image was read also. Subsequently, after calculating the
feature vector of each image, the result was computed by Euclidian distance.
If A and B are two D-distance vectors, using charts [21], the distinction between them
is overcome. Here are the following equations (1) and (2).
Manhattan distance:
(1)
Euclidean distance:
(2)
3.3 Eigenface
Eigen’s face employs the appearance-based method in computer vision to identify the
face of humans. It understands the diversity in the educational variation of photos of
the face, which is afterward used to alter and organize photos [22]. Eigenfaces are the
important section for the distribution of faces. The goal of the Principal part investiga-
tion (PCA) is to make a global error in the preliminary group of pictures and illustrate
this diversity using some variables [23]. This is a dimensionality reduction process that
focuses on getting rid of the required amount of important facial data segments [22].
An eigenvector is one such vector that does not change its way under the associated
direct variation and Eigen’s features are the combination of eigenvectors that chooses
one element from facial picture space [23]. The covariance matrix C is computed, and
by using the following equations (3) [24], the eigenvectors ei and eigenvalues λi are
found out in equation (2):
(3)
(4)
If νi and μi are eigenvectors and eigenvalues of matrix ATA [24] that is shown in equa-
tion (5), then:
(5)
After multiplying both sides together of equation (3) with A, we get the value of equa-
tion (6)
(6)
Applying in equation (6), equation (7) will be,
(7)
The preparation set is altered hooked on a vector P [24], diminished by the mean worth
Ψ and expected by a grid of eigenvectors which are shown in equation (8),
7
(8)
It is evident that after subtracting the collected images into training and testing parts,
the covariant value was computed. Subsequently, eigenvalue and eigenvector were also
calculated before doing projection. Furthermore, all of the required works that had been
done were saved in a specific folder after getting the expected outcomes of projection.
All of the necessary descriptions are added in Fig. 4 to better understand the overall
process.
3
Fig. 4. The Required Steps of Eigenface Technique.
3.4 Speed Up Robust Features (SURF):
SURF [25] is a standard pivotal part that involves a storyline and a descriptor of
intriguing details. The radar finds that the intrigue focuses on the picture, and the de-
scriptor defines the highlights of the emphasis of the conspiracy and constructs the ele-
ment vectors of the focus of the intrigue [26].
1) Interest Point Detection: The SURF indicator was found based on the Hessian
matrix. Assumed a fact X (x, y) in an image I, the Hessian
matrix at X at scale is distinct in this manner [26] and [27]. The equation (9)
is given below.
(9)
Where is the convolution of the Gaussian second order derivative
through the image I at point X, and likewise for and.
8
2) Interest Point Description: The point of focus in a picture is a point in its neigh-
borhood that is special. A two-step strategy is usually used to diagnose and clarify this
point: A. Feature Detectors: where an algorithm that uses an image as input is a feature
detector (extractor) and outputs a set of regions ('local features'). B. Descriptor function:
where a descriptor is computed to a detector-specified picture field. Descriptors are built
by removing rectangular areas about the attention facts. The frames are separation in 4x4
sub-regions [26], [27]. The shape is described by a vector in equation (10) and shown in
Fig. 5.
(10)
In face recognition, SURF characteristics can be derived from photographs through
SURF detectors and descriptors utilizing SIFT functionality. Interest points are first re-
moved from each face picture during pre-processing, such as normalization and histo-
gram equalization. This results in the acquisition of between 30-100 interest points per
photo. The SURF feature vectors [28] of the range of interest points are then determined
to characterize the picture and these feature vectors are normalized to 1. These charac-
teristics are person-specific since each person's picture varies in the amount and position
of points selected by the SURF detector and the characteristics measured by the SURF
descriptor around these points.
Fig. 5. Gaussian second-order partial derivatives and pattern.
4 Experimental Results and Discussion
4.1 Expected Outcomes After Using Eigenface
Some images from the training database are displayed in Fig. 6 and Fig. 7, where all
190 eigenvalues are demonstrated. Every eigenvalue belongs to one eigenvector and
shows us to what extent images from the training database differ from the average im-
age in the same path. It is observed that only 10% of the vectors have considerable
eigenvalues, whereas the remaining vectors consist of eigenvalues almost close to 0.
Eigenvectors consisting of insignificant eigenvalues are needless to consider because
they do not contain critical image data.
9
4
Fig. 6. Training Images.
Fig. 7. Eigenvalues.
In Fig. 8, the first three eigenfaces are displayed as outputs and these images are quite
similar to the input values which had been taken from the explained dataset.
10
Fig. 8. Graphical Representation of Eigenfaces.
The calculated results are displayed based on the different number of principle
elements. Among all of the displayed results, the highest results (approximately 98.5%)
for both Manhattan and Euclidean distances were for 190 eigenvectors. In comparison,
the lowest outcomes were observed for 6 eigenvectors which were about 86.5% and
76.7% respectively. The distance of Manhattan was considerably higher than the
Euclidean distance except for the calculated findings of 20 eigenvectors, which were
almost 0.9% higher, after considering all listed eigenvectors. A sufficient explanation
has been added in Fig. 9.
Euclidean and Manhattan distances have been calculated based on the different types
of images of five persons to carry out the rate of recognition. After evaluating all of
them, the highest predictable distance was noticed for both distances (Euclidean and
Manhattan) where the recognition rate of the 5th person was around 98.5% and 98.8%,
however, the lowest rate of prediction was discovered for 1st person (82.5%, 86.5%).
Besides, Fig. 10 shows that both distances such as Euclidean and Manhattan provide
similar results (only 94%) for the position of 4th.
Fig. 9. The Examined Results of Face Recognition by Eigenface.
11
Fig. 10. The Examined Results of Per Person for Different Number of Images by Eigenface.
Fig. 11. Accuracy based on Image Size.
The corresponding Fig. 11 is illustrated based on various levels of image size. If the
image size is low there is not enough data to process and if too high, it takes too much
time to read and process the image data. So the ideal size is 227*227. For all of these
images, the most predictable accuracy [37-38] which was just over 99.50% was
depicted for 227*227 image size, on the other hand, the last result was witnessed for
the size of 300*300 which accuracy was close to 97%.
4.2 Displayed results of SURF after using different levels of dimensions
Our test results were compared with the SIFT approaches with this proposed method.
Here, the feature vectors of a dimension are indicated by 64 and 128 and dbl refers to
the appropriate size of the given image that was doubled before extracting the feature.
After generating the combined outputs of SURF (64 and 128 dimensions) and SIFT
approaches (128 dimensions), approximately 0.55 threshold values were observed for
the first three consecutive features and around 0.5 threshold results were achieved for
the last three features of SURF and SIFT techniques. The obtainable outcomes have
been clearly shown in the following Fig. 12.
12
Fig. 12. Different ratio of thresholds.
The identification rates on all types of attributes are given in Fig. 13. It is clear that the
identification rate of both SURF-64 and SIFT-128 are the same.
Fig. 13. The comparison of recognition rate between SURF and SIFT features.
The 128-dimensional SURF (SURF-128) attribute vectors are a bit better than SURF-
64 and SIFT-128. In the case of N-dimensional attribute sets, the comparison for
identification exceeds non-doubled attribute sets, but it is weaker for attribute sets
"doubled" with 53 dimensions than "non-doubled" with 64 dimensions. The SURF
profiling algorithm was created to be applied to high-dimensional data. On this
particular dataset that cannot explain with more than 64 dimensions. This is because it
will give rise to higher interest points for a doubled image in contrast to the non-doubled
image, which means 128 dimensions will give higher discrimination data as opposed
to 64 dimensions in the match.
13
5 Conclusion
Modern technology is all about performance and speed [29-30] [39]. Today is the
scientific and technical era. For the present culture, new technology is a great blessing
[31-34]. In every aspect of our lives, we see the application of new technologies [35-
36]. Without science and technology, we cannot conceive about our daily life. This
research focuses on different face recognition methods. To identify human faces, the
eigenface technique is used here. Besides, the SURF process is also demonstrated.
Compared to other approaches and even among the techniques, the accuracy rate of the
ways is seen. It can be seen from the comparison that each approach has its value, which
is dependent on the state of the data. Plans all demonstrate positive progress in the
identification of individual features in any given status.
References
1. F Mahmud, M T Khatun, S T Zuhori, S Afroge, M Aktar dan B Pal, 2015 "Face recognition
using Principle Component Analysis and Linear Discriminant Analysis" Electrical Engineer-
ing and Information Communication Technology (ICEEICT), International Conference on,
pp. 1-4.
2. Mustamin Anggo , La Arapu "Face Recognition Using Fisherface Method" 2nd International
Conference on Statistics, Mathematics, Teaching, and Research, IOP Conf. Series: Journal of
Physics: Conf. Series 1028 (2018) 012119, doi :10.1088/1742-6596/1028/1/012119
3. R. Kaur and E. Himanshi, "Face Recognition using Principal Component Analysis," in IEEE
International Advance Computing Conference, 2015, pp. 585–589.
4. Manoharan, Samuel. "Study On Hermitian Graph Wavelets in Feature Detection." Journal of
Soft Computing Paradigm (JSCP) 1, no. 01 (2019): 24-32.
5. S. Singh and S. V. A. V. Prasad, "Techniques and Challenges of Face Recognition: A Critical
Review," in Procedia Computer Science, 2018, vol. 143, pp. 536–543.
6. Javed Mehedi Shamrat F.M., Allayear S.M., Alam M.F., Jabiullah M.I., Ahmed R. (2019) A
Smart Embedded System Model for the AC Automation with Temperature Prediction. In:
Singh M., Gupta P., Tyagi V., Flusser J., Ören T., Kashyap R. (eds) Advances in Computing
and Data Sciences. ICACDS 2019. Communications in Computer and Information Science,
vol 1046. Springer, Singapore. https://doi.org/10.1007/978-981-13-9942-8_33
7. Smys, S., Joy Iong Zong Chen, and Subarna Shakya. "Survey on Neural Network Architec-
tures with Deep Learning." Journal of Soft Computing Paradigm (JSCP) 2, no. 03 (2020): 186-
194.
8. Md. Razu Ahmed, Md. Asraf Ali, Nasim Ahmed, Md. Fahad Bin Zamal, and F.M. Javed Me-
hedi Shamrat. 2020. The Impact of Software Fault Prediction in Real-World Application: An
Automated Approach for Software Engineering. In Proceedings of 2020 the 6th International
Conference on Computing and Data Engineering (ICCDE 2020). Association for Computing
Machinery, New York, NY, USA, 247–251. DOI:https://doi.org/10.1145/3379247.3379278
9. F. M. Javed Mehedi Shamrat, Z. Tasnim, P. Ghosh, A. Majumder and M. Z. Hasan, "Person-
alization of Job Circular Announcement to Applicants Using Decision Tree Classification Al-
gorithm," 2020 IEEE International Conference for Innovation in Technology (INOCON),
BANGLURU, 2020, pp. 1-5, doi: 10.1109/INOCON50539.2020.9298253.
10. F. M. Javed Mehedi Shamrat, P. Ghosh, M. H. Sadek, M. A. Kazi and S. Shultana, "Imple-
mentation of Machine Learning Algorithms to Detect the Prognosis Rate of Kidney Disease,"
14
2020 IEEE International Conference for Innovation in Technology (INOCON), BANGLURU,
2020, pp. 1-7, doi: 10.1109/INOCON50539.2020.9298026.
11. P. Ghosh et al., "Efficient Prediction of Cardiovascular Disease Using Machine Learning Al-
gorithms with Relief and LASSO Feature Selection Techniques," in IEEE Access, doi:
10.1109/ACCESS.2021.3053759.
12. P. Ghosh, et al. (2020). Expert Model of Cancer Disease Using Supervised Algorithms with a
LASSO Feature Selection Approach. International Journal of Electrical and Computer Engi-
neering, 11.
13. Ibrahim, R.; Zin, Z.M., "Study of automated face recognition system for office door access
control application," Communication Software and Networks (ICCSN), 2011 IEEE 3rd Inter-
national Conference, vol., no., pp.132-136, 27-29 May 2011
14. Pronab Ghosh, Atqiya Abida Anjum, Asif Karim, Masum Shah Junayed, Md. Zahid Hasan,
Khan Md. Hasib, Al Nahian Bin Emran, “A Comparative Study of Different Deep Learning
Model for Recognition of Handwriting Digits,” International Conference on IoT Based Con-
trol Networks and Intelligent Systems (ICICNIS 2020).
15. Junying Gan, Peng Wang, "A novel model for face recognition," System Science and Engi-
neering (ICSSE), 2011 International Conference on , vol., no., pp.482-486, 8-10 June 2011
16. Kshirsagar, V.P.; Baviskar, M.R.; Gaikwad, M.E., "Face recognition using Eigenfaces," Com-
puter Research and Development (ICCRD), 2011 3rd International Conference, vol.2, no.,
pp.302-306, 11-13 March 2011.
17. ORL database. [Online]. Available at: http://www.cl.cam.ac.uk/research/dtg/attarchive/fac-
edatabase.html (Accessed on: 13-12-2020).
18. K Ravi, M Kttswamy "Face Recognition using PCA and Eigenface Approach,” International
Journal and Magazine of Engineering, Management and Research, Vol. 1, Issue No. 10, Oc-
tober, 2014.
19. A.K.M Sazzadur Rahman, F. M. Javed Mehedi Shamrat, Zarrin Tasnim, Joy Roy, Syed Akhter
Hossain "A Comparative Study On Liver Disease Prediction Using Supervised Machine
Learning Algorithms" International Journal of Scientific & Technology Research, Volume 8,
Issue 11, November 2019, ISSN: 2277-8616, pp: 419-422.
20. D. Chakraborty, S.K. Saha, M.A. Bhuiyan, "Face recognition using eigenvector and principle
component analysis", International Journal of Computer Applications, 50(10):42-49 (2015)
21. V. Perlibakas: Distance Measures for PCA-based Face Recognition, Pattern Recognition
Letters, Vol. 25, No. 6, April 2004, pp. 711 – 724.
22. F. M. Javed Mehedi Shamrat, Md. Abu Raihan, A.K.M. Sazzadur Rahman, Imran Mahmud,
Rozina Akter, “An Analysis on Breast Disease Prediction Using Machine Learning Ap-
proaches” International Journal of Scientific & Technology Research, Volume 9, Issue 02,
February 2020, ISSN: 2277-8616, pp: 2450-2455.
23. F. M. Javed Mehedi Shamrat, Zarrin Tasnim, Imran Mahmud, Ms. Nusrat Jahan, Naimul Islam
Nobel, “Application Of K-Means Clustering Algorithm To Determine The Density Of De-
mand Of Different Kinds Of Jobs”, International Journal of Scientific & Technology Research,
Volume 9, Issue 02, February 2020, ISSN: 2277-8616, pp: 2550-2557
24. Marijeta Slavković1, Dubravka Jevtić, "Face Recognition Using Eigenface Approach" Ser-
bian Journal of Electrical Engineering, Vol. 9, No. 1, pp. 121-130, February 2012.
25. H. Bay, A. Ess, T. Tuytelaars, L. Van Gool, "Speeded-up Robust Features (SURF)," Comput.
Vis. Image Underst., Vol. 110 (3), pp. 346-359, 2008.
26. Geng Du, Fei Su, Anni Cai "Face recognition using SURF features", MIPPR 2009: Pattern
Recognition and Computer Vision, Proc. of SPIE Vol. 7496, doi: 10.1117/12.832636 12.
15
27. Donghoon Kim, Rozenn Dahyot, "Face Components Detection using SURF Descriptors and
SVMs" International Machine Vision and Image Processing Conference 2008,
DOI: 10.1109/IMVIP.2008.15.
28. Raúl Cid Carro, Juan-Manuel Ahuactzin Larios, Edmundo Bonilla Huerta, Roberto Morales
Caporal and Federico Ramírez Cruz "Face Recognition Using SURF" ICIC 2015, Part I,
LNCS 9225, pp. 316–326, 2015, DOI: 10.1007/978-3-319-22180-9_31
29. Shamrat F.M.J.M., Nobel N.I., Tasnim Z., Ahmed R. (2020) Implementation of a Smart Em-
bedded System for Passenger Vessel Safety. In: Saha A., Kar N., Deb S. (eds) Advances in
Computational Intelligence, Security and Internet of Things. ICCISIoT 2019. Communica-
tions in Computer and Information Science, vol 1192. Springer, Singapore.
https://doi.org/10.1007/978-981-15-3666-3_29
30. F.M. Javed Mehedi Shamrat, Shaikh Muhammad Allayear and Md. Ismail Jabiullah "Imple-
mentation of a Smart AC Automation System with Room Temperature Prediction", Journal of
the Bangladesh Electronic Society, Volume 18, Issue 1-2, June-December 2018, ISSN: 1816-
1510, pp: 23-32.
31. F.M. Javed Mehedi Shamrat, A.K.M Sazzadur Rahman, Zarrin Tasnim, Syed Akhter Hossain
"An offline and online-based Android application "Travel Help" to assist the travelers visually
and verbally for Outing" International Journal of Scientific & Technology Research, Volume
9, Issue 01, January 2020, ISSN: 2277-8616, pp: 1270-1277.
32. F. M. Javed Mehedi Shamrat, Md Asaduzzaman, Pronab Ghosh, Md Dipu Sultan, and Zarrin
Tasnim, "A Web Based Application for Agriculture: "Smart Farming System"" International
Journal of Emerging Trends in Engineering Research, Volume 8, Issue 06, June 2020, pp:
2309-2320, ISSN: 2347- 3983.
33. F. M. Javed Mehedi Shamrat, Zarrin Tasnim, A.K.M Sazzadur Rahman, Naimul Islam Nobel,
Syed Akhter Hossain "An Effective Implementation of Web Crawling Technology to Retrieve
Data from the World Wide Web (www)" International Journal of Scientific & Technology
Research, Volume 9, Issue 01, January 2020, ISSN: 2277-8616, pp: 1252-1256.
34. F. M. Javed Mehedi Shamrat, Imran Mahmud, A.K.M Sazzadur Rahman, Anup Majumder,
Zarrin Tasnim, Naimul Islam Nobel, "A Smart Automated System Model For Vehicles De-
tection To Maintain Traffic By Image Processing" International Journal of Scientific & Tech-
nology Research, Volume 9, Issue 02, February 2020, ISSN: 2277-8616, pp: 2921-2928.
35. Md. Razu Ahmed, F. M. Javed Mehedi Shamrat, Md. Asraf Ali, Md. Rajib Mia, Mst. Arifa
Khatun "The future of electronic voting system using Block chain" International Journal of
Scientific & Technology Research, Volume 9, Issue 02, February 2020, ISSN: 2277-8616, pp:
4131-4134.
36. F. M. Javed Mehedi Shamrat, Zarrin Tasnim, Naimul Islam Nobel, and Md. Razu Ahmed.
2019. An Automated Embedded Detection and Alarm System for Preventing Accidents of
Passengers Vessel due to Overweight. In Proceedings of the 4th International Conference on
Big Data and Internet of Things (BDIoT'19). Association for Computing Machinery, New
York, NY, USA, Article 35, 1–5. DOI:https://doi.org/10.1145/3372938.3372973.
37. P. Ghosh, M.Z. Hasan, O.A. Dhore, A.A. Mohammad and M. I. Jabiullah, “On the Application
of Machine Learning to Predicting Cancer Outcome”, Proceedings of the International Con-
ference on Electronics and ICT – 2018, organized by Bangladesh Electronics Society (BES),
Dhaka, Bangladesh on 25-26 November, 2018, pp: 60.
38. F.M. Javed Mehedi Shamrat, Md. Asaduzzaman, A.K.M. Sazzadur Rahman, Raja Tariqul Ha-
san Tusher, Zarrin Tasnim “A Comparative Analysis of Parkinson Disease Prediction Using
Machine Learning Approaches” International Journal of Scientific & Technology Research,
Volume 8, Issue 11, November 2019, ISSN: 2277-8616, pp: 2576-2580.
