Major Project Report on Face Mask Detection Using Convolutional Neural Network Submitted By Chandresh Singh (201800624) Parag Ghosh (201800639) In partial fulfilment of requirements for the award of degree in Bachelor of Technology in Computer Science and Engineering (2022) Under the Project Guidance of External Guide Mr. Ashis Pradhan Assistant Professor (SG). Dept. of CSE DEPARTMENT OF COMPUTER SCIENCE AND ENGINEERING SIKKIM MANIPAL INSTITUTE OF TECHNOLOGY (A constituent college of Sikkim Manipal University) MAJITAR, RANGPO, EAST SIKKIM – 737136 Project Completion Certificate This is to certify that the below mentioned student(s) of Sikkim Manipal Institute of Technology have worked under my supervision and guidance from 10 January 2022 to 31 May 2022 and have successfully completed the project entitled “Face Mask Detection Using Convolutional Neural Network” in partial fulfilment of the requirements for the award of Bachelor of Technology in Computer Science and Engineering. University Registration No. Name of Student Course 201800624 Chandresh Singh B.Tech (CSE) 201800639 Parag Ghosh B.Tech (CSE) Mr. Ashis Pradhan, Associate Professor (SG), Sikkim Manipal Institute of Technology, Majitar, East Sikkim -- 737136 I Project Review Certificate This is to certify that the work recorded in this project report entitled “Face Mask Detection using Convolutional Neural Network” has been carried out by Mr. Chandresh Singh (201800624) and Mr. Parag Ghosh (201800639) of Computer Science & Engineering Department of Sikkim Manipal Institute of Technology in partial fulfilment of the requirements for the award of Bachelor of Technology in Computer Science and Engineering. This report has been duly reviewed by the undersigned and recommended for final submission for Major Project Mr. Ashis Pradhan Associate Professor (SG), Department of Computer Science & Engineering Sikkim Manipal Institute of Technology Majitar, East Sikkim – 737136. II 0. Certificate of Acceptance This is to certify that the below mentioned students of Computer Science & Engineering Department of Sikkim Manipal Institute of Technology (SMIT) have worked under the supervision of Mr. Ashis Pradhan, Associate professor(Selection Grade) from 10th January 2022 to 10th June 2022 on the project entitled “Face Mask Detection using Convolutional Neural Network”. The project is hereby accepted by the Department of Computer Science & Engineering, SMIT in partial fulfilment of the requirements for the award of Bachelor of Technology in Computer Science and Engineering. University Registration No Name of Student(s) Project Venue 201800624 Sikkim Manipal Institute of Chandresh Singh Technology 201800639 Parag Ghosh Sikkim Manipal Institute of Technology Dr. Kalpana Sharma Professor & HOD Computer Science & Engineering Department Sikkim Manipal Institute of Technology Majitar, Sikkim - 737136 III Declaration We the undersigned, hereby declare that the work recorded in this project report entitled “Face Mask Detection using Convolutional Neural Network” in partial fulfilment for the requirements of award of B.Tech in Computer Science & Engineering from Sikkim Manipal Institute of Technology (A constituent college of Sikkim Manipal University) is a faithful and bonafide project work carried out at Sikkim Manipal Institute of Technology under the supervision and guidance of Mr. Ashis Pradhan , Associate Professor(SG). The results of this investigation reported in this project have so far not been reported for any other Degree / Diploma or any other Technical forum. The assistance and help received during the course of the investigation have been duly acknowledged. Chandresh Singh (201800624) Parag Ghosh (201800639) IV Acknowledgement We would like to convey my heartful thanks to our Internal guide Mr. Ashis Pradhan, Associate Professor(SG) without whom the completion of this project would not have been possible. We pay deep sense of gratitude to Prof. (Dr.) Kalpana Sharma, H.O.D, Computer Science & Engineering Department, Sikkim Manipal Institute of Technology for giving me the opportunity to work on this project and provided all support required. We would like to express my humble gratitude to Mr. Biswaraj Sen, Associate Professor, Mr. Santanu Kumar Misra, Associate Professor, Mr. Saurav Paul, Assistant Professor-I, and Mrs Chitrapriya N., Assistant Professor-I, Project Coordinators, Computer Science and Engineering Department, Sikkim Manipal Institute of Technology for their unlisted encouragement and their timely support and guidance till the completion of the project work. Chandresh Singh (201800624) Parag Ghosh (201800639) DOCUMENT CONTROL SHEET V DOCUMENT CONTROL SHEET CSE/Major Project/B.Tech/In-House/Group- 1 Report No 2 Title of the Report Face Mask Detector 3 Type of Report Technical 4 Author(s) 1 /2022 Chandresh Singh Parag Ghosh 5 Organizing Unit Sikkim Manipal Institute of Technology 6 Language of the Document English The project aims to develop a model that”can accurately detect masks over the 7 Abstract face in “public areas”(such as “airports, railway stations”, crowded marketplaces) to reduce the spread of Coronavirus. Therefore, contributing to public” health. 8 Security Classification General VI LIST OF CONTENTS SL NO. TITLE PAGE NO. 0 Abstract 1 Introduction 1–5 1.1 General Overview of the Problem 1-2 1.2 Literature survey 3-4 1.3 Problem definition 5 1.4 Software Requirement and Specification 6 1.5 Dataset 7 1.5 Solution Strategy 8 2 3 0 Design 9 -12 2.1 Flow Chart 9-10 2.2 Sequence Diagram 11 2.3 Activity Diagram 12 Implementation 13-22 3.1 Keras 13 3.2 Tensorflow 13 3.3 Pseudo code for Phase:1 14 3.4 Pseudo code for Phase:2 15 3.5 Pseudo code for Phase:3 15 3.6 Scene Builder 17 3.7 Video Capturing 18 3.8 Mask using CNN 20-21 3.9 Accuracy 22 3 Results and Discussion 23-24 4 Summary And Conclusion 25 5 Limitations and Future scope 26 6 Gantt Chart 27 References 28 Plagiarism Report VII LIST OF FIGURES SL TITLE NO. PAGE NO. 1 Block Diagram (phase 1,2,3) 9-10 2 Sequence Diagram 11 3 Activity Diagram 12 4 Face Detection using Haar 16 5 User Interface for Face Detection 17 6 BGR color in OpenCV 19 7 CNN model hidden layer for Face Mask detector 20 8 Formula to calculate accuracy for the model 21 9 Graph for training accuracy vs validation accuracy 21 10 Graph for training loss vs validation loss 22 11 Graph for training loss vs validation loss 22 12 Mask detection on a face with no mask results in a 100 % 23 No-mask. 13 Person wearing mask inappropriately resulting in mask 23 coverage as 66.97 14 Confusion matrix for face mask detection 24 15 Gantt Chart 28 VIII LIST OF TABLES TABLE TABLE NAME NO. 1.1 PAGE NO. Literature Survey 3-4 IX ABSTRACT With the reopening of places and businesses under lockdowns, health institutions are suggesting Face masks as “essential measures” to keep citizen “safe when” venturing outdoors. To mandate the use of face masks, it becomes essentials as a country’s citizen to help “enforce individuals to apply mask before exposure to public places”. Here HAAR-CASCADE algorithm can be used for feature detection in an image. These” classifiers “result a high recognition” rate with range of “expressions”, efficient” feature “selection” and “low assortment” of “false positive” features. Cascade classifiers like “HAAR uses” only “200 features “out of 6000 features” to yield recognition of about 85-95%. The project aims to develop a “technique that”can accurately”detect masks over the face in “public areas” (such as “airports, railway stations”, crowded marketplaces) to reduce the spread of Coronavirus . Therefore, contributing to public” health. 0 1. INTRODUCTION 1.1 General Overview of the Problem COVID-19, a novel virus, first detected in Wuhan in 2019.Which later become a global public health issue. This disease is wreaking havoc on the world's poorer economies. Severe Acute Respiratory Syndrome Coronavirus 2 is a once-in-a-century respiratory viral illness (SARS-CoV-2). The pandemic is wreaking havoc on society and economies all across the world, resulting in a global health emergency. It has been a major healthcare concern throughout the world, particularly in the third wave. As a result of the outbreak, several businesses have closed their doors. Furthermore, due to their enormous impact on people's daily lives, numerous areas such as maintenance projects and infrastructure building have not been interrupted.” The virus had rapidly spread to most countries throughout the planet as of April 2021.According to the most recent WHO figures, there have been 152,543,452 confirmed cases and 3,198,528 deaths. Corona-virus infection is transmitted primarily through respiratory droplets produced when people breathe, talk, cough, or sneeze, with a common droplet size of 0.5-1.0m, according to the Centres for Disease Control and Prevention (CDC), but aerosol emission increases when humans speak or shout loudly.” Because present regulatory measures are unable to prevent COVID-19 from spreading rapidly, most world governments have proposed a variety of solutions, including imprisonment and lockdowns. However, in addition to the public goods game, game-theoretic scenarios can be used to investigate COVID-19 management inefficiency. Some academics have concentrated on governments' reluctance to impose onerous but necessary viral containment measures (such as stay-at-home orders and lockdowns), as well as noncooperation for reasons other than free riding. For example, because tight stay-athome measures can have a significant impact on people's livelihoods, the cost of staying at home (combined with lockdown weariness) may end up outweighing the risk of infection from venturing out, according to authors in.” 1 When vaccinations become accessible at the end of 2020, scientists predicted that vaccine adoption programs will outstrip other factors including vaccine efficacy and isolation processes. Using social network analysis, and agent-based modelling, the scientists argued that "demographics, physical location, the level of interaction, the health of the vaccine,” epidemic parameters, and perceptions about the vaccine being introduced" would influence people's vaccination decisions, while "epidemic parameters, the nature of the vaccine being introduced, logistics, the management of human resources" would influence government decisions. To summarise, COVID-19 management necessitated an awareness of a range of elements that calibrated payoffs and changed individual and governmental behaviours toward safety. True, COVID-19 is a global epidemic that affects a variety of domains. Nonetheless, it paved the way for computer science researchers. A variety of study areas, including developing new COVID-19 automatic detection algorithms, and recognising people wearing or not wearing masks. Using dataset from Because there have been some flaws and delays in the early laboratory testing, researchers have concentrated on various approaches As a result, combining modern AI techniques with chest radiological imaging can lead to a more accurate detection of the COVID19 and aid in addressing the problem of specialised physician shortages in remote areas. 2 1.2 Literature Survey YEA R PAPER DETAILS AUTHO R FINDING PROJECT RELEVAN CE 2022 A real time face mask detection Hiten system using convolutional neural Goyal network Karanve Publication Date-25/2/22 er Sidana The model uses CNN as backbone architectur e which can be implement ed for any image to detect presence of a mask. The Image is then fed to the face detection model first to detect all the faces in a provided image The Paper also discuss about face mask detection system for static images to identify if a person picture has a mask on or not. Multimedia Tools and Applications, Charanje 81(11) et Singh Pp:7-12 3 2017 Face Detection and Tracking Using OpenCV Publication Date-31/07/2019 International_Conference_on_Electr onics, Communication_and Aerospace Technology Pp: 474-478 Kruti Goyal, Kartikey Agarwal, Rishi Kumar The paper proved helpful in image processing, and feature tracking of the face It talks about the Haarclassifier Table 1.1: Literature Survey 4 The paper was helpful in segmentation of an image for feature extraction using the haar classifier and storing the interested region into an array 1.3 Problem Definition Government of India has 2 clear mandates for public health and security of citizens to fight against Covid-Masks and Vaccines. The later having a coverage of 62% of adults being vaccinated. Many counters for people not wearing masks have been implanted including fines and even jail under section 188.People keep finding ways to break the law when no-one is watching. The process of keeping in check if the citizens are wearing masks or not is particularly manually taxing in crowded places like public institutions and “private institutions. As very small proportion of the people wear masks most amongst them still tend to use it not for safety rather than for symbol to show that they have it. Wearing mask by not properly covering their mouth and face. Referred pdf - “The later having a coverage of 62% of adults being vaccinated” https://www.mohfw.gov.in/pdf/GuidelinesforCOVID19VaccinationofChildrenbetwee n15to18yearsandPrecautionDosetoHCWsFLWs&60populationwithcomorbidities.pdf 5 1.4 Software Requirement and Specification A. Hardware Specification of Developing Environment • Laptop/computer with minimum RAM: - 8 GB • Hard Disk: 360 GB or more • Processor: Intel Pentium i3 or above • Graphics Card: 2GB or more B. Software Specification • OS: Windows 8 or above, Linux, Mac • Python (version 3.10) • Scene Builder (UI controller), Pycharm, Juypter Notebook • Libraries: OpenCV, Tensorflow, Keras, Pytorch 6 1.5 Dataset The dataset used in this research was collected in picture formats of JPEG Figure 1 exhibits the sample of the dataset. It was a mixture of different open-source datasets and images, including dataset for Face Mask Detection by techhoney and dataset for PYImageSearch_reader by Prajna Bhandary. As a result, there were different varieties of images with variations insize and resolution. All the photos were from open-source resources, out of which some resemble real-world scenarios, and others were artificially created to put a mask on the face. Major Facial landmarks include the eyes, nose, chin, lips, and eyebrows. This intelligently creates a dataset by forming a mask on a non-masked image. Finally, the dataset was divided into two classes or labels. These were ‘with mask’ and ‘without mask’, and together, the images were curated, aggregating to around 4000 images. 7 1.6 Solution strategy Phase 1: Data pre-processing 1.1: Load the images from the dataset 1.2: Label images into classes ([0] for the class containing mask images, [1] for the class containing without-mask images) 1.3: Spilt dataset into test and train (80:20 as the total images in dataset < 10,000 images) 1.4: Apply data augmentation (resize, crop, shift-to-right, shift-to-left, rotate 90 degrees) Phase 2: CNN model training for Mask 2.1: Use ReLu as Activation function in Conv2D layer 2.2: Add Maxpool2D layer 2.3: Add flatten layer to the model 2.4: Use model. eval function to add the dropout layer in the model 2.5: Use Softmax as Activation function in the dense layer which use probability to generate binary output. 2.6: Use model. train to train the model on training dataset. Phase 3: Applying Mask model on face detector 3.1: Load the saved mask CNN model 3.2: Convert input image from BGR to Grayscale 3.3: Import frontal_face.xml model for face detection 3.4: Use multiscale function (returns value in an array [ x ,y ,w , h] that can be used to draw a rectangle around the detected face). 3.5: Input the resulting array into the masked model to get binary output of 0(mask) and 1(No-mask). 8 2. DESIGN 2.1 Flow charts depicting different phases for the model of face mask detector Fig 1.1: Flow Chart for Phase 1 Fig 1.2: Flow Chart for phase 2 9 Fig 1.3: Flow Chart for phase 3 10 2.2 Sequence Diagram Fig 2: Sequence Diagram A sequence diagram is a type of interaction diagram because it describes how and in what order a group of objects works together. These diagrams are used by software developers and business professionals to understand requirements for a new system or to document an existing process. Sequence diagrams are sometimes known as event diagrams or event scenarios. Sequence diagrams can be useful references for businesses and other organizations. ●Represent the details of a UML use case. ●Model the logic of a sophisticated procedure, function, or operation. ●See how objects and components interact with each other to complete a process. ● Plan and understand the detailed functionality of an existing or future scenario.” 11 2.3 Activity Diagram Fig 4: Activity Diagram As the input is taken from the camera, frame rate and a small of delay of 1 ms is set. A multi-scale model is used for feature extraction using haar based edge classifier to detect face in an image. When the face is not detected, we capture the image again and call multi-scale until a face is found. But when a face is found we import the mask model to detect mask. Which at the end uses SoftMax layer to return a output using probability function resulting in an output of mask and mask-less. 12 3. IMPLEMENTATION 3.1 Keras Keras is an API designed for human beings, not machines. Keras follows best practices for reducing cognitive load: it offers consistent & simple APIs, it minimizes the number of user actions required for common use cases, and it provides clear & actionable error messages. Keras is a minimalist Python library for deep learning that can run on top of Theano or Tensor Flow. It was developed to make implementing deep learning models as fast and easy as possible for research and development Keras was developed and maintained by François Chollet, a Google engineer using four guiding principles: Modularity: A model can be understood as a sequence or a graph alone. All the concerns of a deep learning model are discrete components that can be combined in arbitrary ways. Minimalism: The library provides just enough to achieve an outcome, no frills and maximizing readability. Extensibility : New components are intentionally easy to add and use within the framework, intended for researchers to trial and explore new ideas. Python: No separate model files with custom file formats . Everything is native Python . Keras is designed for minimalism and modularity allowing you to very quickly define deep learning models and run them on top of a Theano or TensorFlow backend. 3.2 Tensorflow TensorFlow is a free and open-source software library for dataflow and differentiable programming across a range of tasks. It is a symbolic math library, and is also used for machine learning applications such as neural networks. It is used for both research and production at Google, TensorFlow is Google Brain's second-generation system. While the reference implementation runs on single devices, TensorFlow can run on multiple CPUs and GPUs (with optional CUDA and SYCL extensions for generalpurpose computing on graphics processing units). 13 3.3 Algorithm for phase 1-: Data pre-processing 3.3.1. Load the images from dataset Step:1 Start Step:2 Set experiment path to the path of a newly created folder Step:3 Set data_path to the path where dataset is stored Step:4 Set file_path with os.join with experimental_path and data_path Step:5 Read files from file_path Step:6 For loop starts for file in file_path 6.1 check if file is from mask folder 6.1.1 6.1 append file to file_path with its name changed to 0 if not append file to file_path with its name changed to 1. Step:7 End 3.3.2 Split the dataset into test and train Step:1 Start Step:2 Set file_name to empty Step:3 Read files from file_path where the combined dataset is stored Step:4 Set train_size to 0.8 with random_state being any positive number Step:5 Set test_size to 0.2 with random_state being same to that of the train dataset. Step:6 End 3.3.3 Apply data augmentation Step:1 Start Step:2 Perform resize () to make all that training images the same size Step:3 Convert image to tensor Step:4 Perform Normalize on the image change to the range of pixel intensity value Step:5 Repeat step 1-4 for test images Step: 6 Stop 14 3.4 Algorithm for phase 2-: CNN model training for Mask 3.4.1 Use model.train() to train model on training dataset. Step:1 Start Step:2 Set the model in train mode Step:3 Make an array for loss[] Step:4 For loop start for batch in train dataset Step:5 train step(defines the training step using gradient descent) Step:6 clear gradients by setting optimizer_grad to zero. Step:7 backward loss to calculate gradient Step:8 update the parameter using append loss. Step:9 Stop 3.5 Algorithm for phase 3-: Mask model on face Detector Step:1 Start Step:2 Load the mask model Step:3 Use image processing function(imgproc) to convert BGR images into Grayscale Step:4 Import frontal_face.xml classifier Step:5 Find centre of the image Step:6 Use multiscale function(to return value in matofRect[x,u,w,h]. Step:6 Use matofRect to draw a rectangle around the face Step:7 Pass matofRect to the mask model to get probability function output of 0 and 1. Step:8 Use value of output to print (0 for mask) and (1 for without mask) Step:9 Stop 15 Fig 4: Face detection using Haar 3.6 Scene Builder Fig 5: User Interface for face detection 16 The need to control the button push and the refreshment of the image view. To do so we have to create a reference between the GUI components, and a variable used in our controller class: The @FXML tag means that we are linking our variable to an element of the fxml file and the value used to declare the variable has to equal to the id set for that specific element. 3.7 Video Capturing All the functionalities required for video manipulation is integrated in the Video Capture class. In “case of a video file there is a frame rate specifying just how long is between two frames. While for the video cameras usually there is a limit of just how many frames they can digitalize per second. In our case we set as frame rate 33 frames per sec.” 17 BGR color order in OpenCV OpenCV “loads the color images in reverse order and uses the BGR color format instead” of the RGB. We can see the order of the channels in the following diagram: Fig 6: BGR color in OpenCV 18 Now we are going to convert our image from BGR to Grayscale format. OpenCV has a really nice function to do this kind of transformations: “ cvtColor takes as arguments: a source image (frame) a destination image (frame), in which we will save the converted image. an additional parameter that indicates what kind of transformation will be performed. In this case we use COLOR_BGR2GRAY ( BGR default channel order in case of colour images).” This encodes an image into a memory buffer. The function compresses the image and stores it in the memory buffer that is resized to fit the result.” It takes three parameters: (“. Png”) format for the output (frame) Image to be written. (buffer) Output buffer resized to fit the compressed image. 3.8 Mask Detector Fig 7: CNN model hidden layers for Face Mask Detector Fully Connected Layer: It's important to add a fully connected after “convolution, non-linear, and pooling layers functions are done. This layer receives output data from all the convolution network layers. Fully connected layer is used at the ends of a network making an N- 19 dimensional vector, wherein N means the number of classes through which the model chooses the needed class.” Layers In CNN Model: Conv2D Flatten () MaxPooling2D Dense Dropout 1. Conv2D: This. The Rectified Linear Unit (ReLu) function will return the data input if it really is positive else, it will output zero. 2. Flatten (): It's used to blend all of the levels into a single one-dimensional layer. 3. MaxPooling2D: It is utilized with a 2*2 pool or filter size. 4. Dense: Soft max is the activation function used here, and it produces a vector with two probability distribution values. 5. Dropout: It's utilised to keep the model from fitting too tightly, or in other words, from overfitting. 20 3.9 Accuracy It tells about the performance or how the model is performing in general Quite useful when all the classes are of the equal importance Calculate in the ratio between the number of correct predictions to the total number of” predictions Fig:8-Formula to calculate accuracy of the model In this model the accuracy is calculated using Scikit-learn, dividing the sum of true positives and true negatives Over all the values of confusion matrix. Fig 9: “Graph for training accuracy vs validation accuracy Fig 10: Graph for Training loss vs Validation loss 21 3.10 Confusion Matrix Fig 11: Confusion matrix for Face Mask detector Formulas used - Precision=Tp/(Tp+Fp)=2709/(2709+81)=0.97 Recall=Tp/(Tp+Fn)=2709/(2709+135)=0.95 Accuracy=Tp+Tn/(Tp+Tn+Fp+Fn) =2709+903/(2709+903+81+135)=0.94 Misclssification=Fp+Fn/(Tp+Tn+Fp+Fn) =81+135/(2709+903+81+135)=0.056 Specificity=Tn/(Tn+Fp)=903/(903+81)=0.92 Source-https://towardsdatascience.com/taking-the-confusion-out-of-confusionmatrices-c1ce054b3d3e 22 5. RESULTS AND DISCUSSION Fig 12: Mask detection on a face with no mask results in a 100 %No-mask. Fig 13: Person wearing mask inappropriately resulting in mask coverage as 66.97%. 23 Fig 14: Person wearing mask appropriately resulting in mask coverage as 100.00%. 24 4. SUMMARY AND CONCLUSION 4.1 Summary The proposed solution for the problem definition is developed and the test cases have been tested and the applications function properly. In this work, a deep learning-based approach for detecting masks over faces in public places to curtail the community spread of Coronavirus is presented. The proposed technique efficiently handles occlusions in dense situations by making use of an ensemble of single and two-stage detectors at the pre-processing level. The ensemble approach not only helps in achieving high accuracy but also improves detection speed considerably. Furthermore, the application of transfer learning on pre-trained models with extensive experimentation over an unbiased dataset resulted in a highly robust and low-cost system. The identity detection of faces, violating the mask norms further, increases the utility of the system for public benefits. 25 5. LIMITATIONS AND FUTURE SCOPE 5.1 Limitation of the project The current limitations of the project are as follows: Less effective against varying angle of the face in the camera, as most data set available that are used in training the model are front facing with clear feature demarcation, this would prove to be hard to determine if a person is wearing a mask or not.” Lack of features-> as the input image or data does not have proper facial features or the features are altered like eyes are covered with sunglasses, in these types of cases, the model would not be able to tell between masked or mask less.” 5.2 Future Scope of the project The future scope of the Face Mask Detector: ● Model can be integrated with any high-resolution video surveillance devices and not limit the to mask detection. ● The model can be extended to detect facial landmarks with a facemask for biometric purposes. 26 6. GANTT CHART Time Frame Activity January February March (2022) (2022) (2022) Literature Survey Problem Definition Design Implementation Testing and Analysis Documentation LEGEND Proposed Activity Activity Achieved Fig 15: Gantt chart 27 April (2022) May (2022) June (2022) REFERENCES [1] M. R. Bhuiyan, S. A. Khushbu and M. S. Islam, "A Deep Learning Based Assistive System to Classify COVID-19 Face Mask for Human Safety with YOLOv3," 2020 11th International Conference on Computing, Communication and Networking Technologies (ICCCNT) Accessed:March 23,2022 [2] M. M. Rahman, M. M. H. Manik, M. M. Islam, Mahmud and J. -H. Kim, "An Automated System to Limit COVID-19 Using Facial Mask Detection in Smart City Network," 2020 IEEE International IOT, Electronics and Mechatronics Conference (IEMTRONICS), 2020. Accessed:February 12,2002 [3] Y. Sun, Y. Chen, X. Wang, and X. Tang, “Deep learning face representation by joint identification- verification,” in Advances in neural information processing systems, 2014, A. Khosla, M. Bernstein, A. C. Berg, and L. Fei-Fei, “ImageNet Large Scale Visual Recognition Challenge,” 2014.Accessed March 5,2022 [4] F. S. Samaria and A. C. Harter, “Parameterisation of a stochastic model for human face identification,” in Applications of Computer Vision, 1994., Proceedings of the Second IEEE Workshop on, pp. 138–142, IEEE, 1994.Accessed April,2022 [5] D. Yi, Z. Lei, S. Liao, and S. Z. Li, “Learning face representation from scratch,” CoRR abs/1411.7923, 2014 Accessed February 23,2022 28