You are on page 1of 8

MSc.

Research Proposal
Master Program < electrical and electronics >
<Feature selection on face recognition using particle swarm
optimization (pso) based on image processing>

<Yasir Saeed Ismael>

yasser_barzani@yahoo.com

Department of<Electrical and Electronics Engineering>


Feature selection on face recognition using particle swarm optimization (pso) based
on image processing

1. Introduction:

Automatic Face Detection (AFR) is a field with a high practical potential for both
business and law enforcement applications, and is thus quite an active field in computer
sciences. Despite the fact that the extensive research in the field of face recognition has
not been given pause since the last three decades, significant progresses were only
achieved in recent years. Challenges such as variety at the amount and angles of light
present in the target medium, the potential variations on the position and angle of the
target face, and potential elements that could partially cover the face still persist [1] and
make the research increasingly difficult. The main reason for this is that most of the
available methods for face recognition are based on visual images [2]. Therefore, the
selection of unique features and distinctive elements of facial images that are independent
of the above challenges, and then adapting these to the features in the available data for
comparisons and evaluations, can be a suitable solution to address these problems. The
success of any face detection method is strongly dependent on the accurate detection of
these features and then making accurate comparisons with the correct feature data subset,
so that factors like variations in light intensity, the angle of the face, and minor features in
the face, do not disrupt the correct diagnosis [3].

1.2. Statement of the problem

In this research, the goal is to implement the mechanism that will identify and
recognize the faces automatically with the help of the image received from the person. A
pre-processing will initially be performed on the image taken. This preliminary process
will involve determining the face area and breaking it down for subsequent processing.
Then, with the help of the image processing engine, the image will be prepared for
feature extraction. This step needs to be performed with high precision so that
environmental factors such as light intensity, minor changes in facial appearance, and
angle of image, etc. do not have strong influence on the output. After the image is
processed correctly, the image properties will be extracted, and the best subset of these
features will be determined using the particle swarm optimization algorithm. Finally,
with the help of a set of selected features, the face will be detected and identified.

1.3. Importance of the study

The world is ever progressing and all kinds of activities are moving towards
automation. Robots play a more significant role in the everyday activities of humans and
the industrial cycles, and computer processors are much stronger and influential than
before. In line with these developments, automatic identification of people has also
become an important functionality. The subject of automatic human recognition by
computers has been the subject of numerous researches over the past two to three
decades, and is a field still open and evolving. Despite considerable advances in this
field, a method without or with minimal margin of error has still not been achieved. This
is due to the fact that even the slightest error in diagnosis during an auto-diagnostic
operation can cause very strong influence over the result.

1.4. Research purposes

       1. Use of image processing capabilities before extracting features from facial images
to enhance the accuracy of detection

       2. Use of PSO optimization method to select the best set of image attributes

1.5. Research questions

1. Is it possible to make facial recognition more accurate by using image processing


techniques along that extract facial features along with optimization algorithms?

2. Is the PSO algorithm suitable for use in solving the optimization problem in Face
Detection Project? If the answer is yes, is it the best way?
3. According to the previous research, which target function is suitable for solving the
optimization problem?

1.6. Research hypothesis

In this research, different thermal images are assumed to be available from different faces
of different angles, possibly with noise. Here, there is no limitation on the age and gender
of the people.
2. Literature review

In [2], the author presents an algorithm for selecting facial features by solving an
optimization problem called "Particle Swarm Optimization (PSO)" in order to identify
faces of individuals. In the first step, the faces in the image are processed and the area of
the whole face is detected. The second step is to use a fixed-scale feature conversion
technique (SIFT) to extract the feature from the face. The features obtained with SIFT are
invariant to the rotation or change of photo scale, and this is the strength of this method.
Some unwanted features, such as noise or non-related features, and features that can be
expressed by the main features, are eliminated during this phase. In other words, optimal
features are chosen for better accuracy in detection. To select this refined feature set, the
optimization features of the PSO algorithm are essential. The function that solves this
optimization problem minimizes the classification gap. The accuracy of the results was
reported for SIFT technique and PSO pattern for images as 98.61% and 90.28%
respectively, for all facial images.

In the reference paper [3], PSO optimization algorithm was selected as an optimization
problem solving tool to determine the subset of optimal features among all extracted
features. This paper has been characterized by two extraction techniques: discrete cosine
transform (DCT) and discrete wavelet transform (DWT). The objective function that is
considered in solving the optimization problem is the maximization of the class
separation. This article shows that the PSO algorithm has yielded results with a minimum
number of attributes.

In the reference [4], the honeybee optimization algorithm was used to determine the
optimal image characteristics for feature detection. In the proposed algorithm, the
classification quality and vector length of the selected features were obtained as
information determined by the bee algorithm. The advantage of this was to obtain the
optimal subset of the selected features without having the prior information about the
properties.

In reference [5], a different face recognition method was proposed based on the DWT /
DCT composition to select the face image features. This method reduces the dimensions
of the image of the original face, but retains the characteristics of the data in the space of
the features. Subsequently, the vector support vector (SVM) is used to classify feature
vectors based on minimizing structural risk.

In the reference article [6], the author presented two methods for extracting the
characteristics. The first method is to extract the two-dimensional DWT-based feature in
the frequency domain (DDFFE). The second is to optimize the binary particle swarm at
the threshold (ThBPSO). The DDFFE method combines three approaches to DFT, DWT
and DCT. In this method, fixed-mode properties, transitions and image brightness are
used for more effective extraction. Furthermore, ThBPSO method is based on the return
of selected features and utilization of them in determining the optimal subset of image
attributes. The advantage of the results of this paper is an increase of accuracy to 98.2%,
and reduction of the maximum number of optimal features required for detection.

In [7], the particle swarm optimization algorithm is used with a perspective on


optimization for face detection. Initially, localized binary patterns, which include vertical
and horizontal pixel values, are determined. These determine the initial range of the face
properties. Then, a PSO type solver embedded with the concept of micro-genetic
algorithm (mGA) is used to optimize the features. In this method, in order to avoid the
early convergence of the PSO algorithm which would cause early congestion, several
factors -including the use of a secondary population of particles smaller than the original
population, the use of the new particle speed update method, the use of An in-depth local
search method for face features, and interaction between local searches and general
searches in search space- was exploited. Despite the complexity of the method, the results
show a higher accuracy than the previous ones.
3. Method (data collection and analysis method):

In this research, validated journals and books will be used as reference points. The
software used in the image processing, feature extraction, and optimization problem
solving software will be the MATLAB software.

3.1. Location (site) / Range / Statistical Society and Sample Size

This research requires the image of a person's face, which, according to valid
papers that have been active in this field, will be sufficient in providing the necessary
data.

3.2. Application of results / exploiters:

All administrative and industrial bodies, universities, and even schools dealing
with the presence and absence of employees and students can use the results of this
research. They also need a system in security centers to identify people.
References

[1] Ghiass, Reza Shoja, et al. "Infrared face recognition: a literature review." The 2013
International Joint Conference on Neural Networks (IJCNN). IEEE, 2013.

[2] Seal, Ayan, et al. "Feature Selection using Particle Swarm Optimization for Thermal
Face Recognition." Applied Computation and Security Systems. Springer, New Delhi,
2015. 25-35.

[3] Ramadan, Rabab M., and Rehab F. Abdel-Kader. "Face recognition using particle
swarm optimization-based selected features." International Journal of Signal Processing,
Image Processing and Pattern Recognition 2.2 (2009): 51-65.

[4] Kanan, Hamidreza Rashidy, Karim Faez, and Mehdi Hosseinzadeh. "Face recognition
system using ant colony optimization-based selected features." 2007 IEEE Symposium on
Computational Intelligence in Security and Defense Applications. IEEE, 2007.

[5] Yu, Ming, Gang Yan, and Qing-Wen Zhu. "New face recognition method based on
dwt/dct combined feature selection." 2006 International Conference on Machine
Learning and Cybernetics. IEEE, 2006.

[6] Krisshna, NL Ajit, et al. "Face recognition using transform domain feature extraction
and PSO-based feature selection." Applied Soft Computing 22 (2014): 141-161.

[7] Mistry, Kamlesh, et al. "A micro-GA embedded PSO feature selection approach to
intelligent facial emotion recognition." IEEE transactions on cybernetics 47.6 (2017):
1496-1509.

You might also like