Masked Face Recognition Using Convolutional Neural Network

Masked Face Recognition Using Convolutional Neural Network

Table of Contents


Recognition from faces is a popular and significant technology in recent years. Face alterations and the presence of different masks make it too much challenging. In the real world, when a person is uncooperative with the systems such as in video surveillance then masking is further common scenarios. For these masks, current face recognition performance degrades. An abundant number of researches work has been performed for recognizing faces under different conditions like changing pose or illumination, degraded images, etc. Still, difficulties created by masks are usually disregarded. The primary concern to this work is about facial masks, and especially to enhance the recognition accuracy of different masked faces. A feasible approach has been proposed that consists of first detecting the facial regions. The occluded face detection problem has been approached using Multi-Task Cascaded Convolutional Neural Network (MTCNN). Then facial features extraction is performed using the Google FaceNet embedding model. And finally, the classification task has been performed by Support Vector Machine (SVM). Experiments signify that this mentioned approach gives a remarkable performance on masked face recognition. Besides, its performance has been also evaluated within excessive facial masks and found attractive outcomes. Finally, a correlative study also made here for a better understanding.

Author Keywords

  • Face detection,
  • MTCNN,
  • FaceNet,
  • Face embedding,
  • Masked face,
  • SVM,
  • Face recognition

IEEE Keywords

  • Face,
  • Face recognition,
  • Databases,
  • Face detection,
  • Support vector machines,
  • Machine learning,
  • Feature extraction


Face recognition is a promising area of applied computer vision [1]. This technique is used to recognize a face or identify a person automatically from given images. In our daily life activities like, in a passport checking, smart door, access control, voter verification, a criminal investigation, and many other purposes face recognition is widely used to authenticate a person correctly and automatically. Face recognition has gained much attention as a unique, reliable biometric recognition technology that makes it most popular than any other biometric technique likes password, pin, fingerprint, etc. Many of the governments across the world also interested in the face recognition system to secure public places such as parks, airports, bus stations, and railway stations, etc. Face recognition is one of the well-studied real-life problems. Excellent progress has been done against face recognition technology throughout the last years.

By the fast improvement and expansion in machine learning techniques, the dilemma of face recognition appears to be fully addressed. During current years, deep learning has obtained numerous breakthroughs in several computer vision areas, such as object detection, object classification, object segmentation, and of course, face detection and verification [2, 3]. The previous face detection and verification algorithms need to manually design features, where deep learning methods do not require manual design. From training images, CNN can learn valuable features automatically. Recently, due to the success in detection and recognition problems, CNNs gained its popularity. CNN’s successively applies convolution filters and they are accompanied by many non-linear activation functions. Before into this area uses a few convolutions filters supported by the average or sum pooling at the image [4]. Recently a more substantial quantity of filters is used which are pre-trained at huge datasets [5, 6, 7]. These methods are useful for detecting faces in various adjustments and poses [8]. The recognition accuracy of the traditional Eigenface algorithm at LFW is barely 60%. Where the recognition accuracy of the most advanced deep learning algorithm holds 99.63%. The acceptance rate is higher than the normal human eye because the human eye acceptance rate is 99.25% [9]. Multiple international projects have been effectively applied deep learning methods like FaceNet, DeepFace, and many more for face recognition. Within those algorithms, the leading accuracy holds FaceNet and at the LFW dataset, it reached a 99.63% accuracy rate, which is higher than the normal human eye [9].

Although while handling unrestrained circumstances into which image degradation, changes in facial pose, occlusions, and other constraints normally occurs, popular systems still suffer. An individual can be disguised his identity by face alterations or using different altered physical attributes. For several deliberate or unintentional reasons, facial occlusions may have occurred. As an example, dacoit, hooligans, and offenders use sunglasses or scarves to block their faces for being unrecognized. Many people use a hood, make beards as religious faith or social tradition. Other origins regarding occlusions involve medical masks, caps, mustaches, makeup, etc. Sometimes faces are not clear, because many obstacles can be in front of the face. These obstacles can be a different object, glasses, scarves, caps, or some occlusion on the face. Face occlusion problems can be classified into three categories in real-world: occlusion of facial landmarks, occluded by different objects, and occluded by faces [6]. Facial landmark occlusion involves using masks and sunglasses. Occluded by faces is a complicated circumstance where misrecognized several faces towards one face or hardly detect a portion of a face region. When occlusion caused by any objects, normally most of the faces will remain hidden.

If mask analysis does not particularly take into consideration, then it can change the performance seriously most of the sophisticated face recognition methods. The key features to identify a person are decreasing by using various sorts of masks or occlusions. Fewer numbers of facial features in the masked face cause difficulties than other normal face recognition techniques [10]. Therefore, the accuracy rate of recognition is decreasing. For disguising identities, terrorists and criminals are covered their faces with the mask. That’s why the masked face is being one of the majors concerning factors within the domain of face recognition. On the other hand, the usage of a deep learning network is more challenging because the quantity of training data is not sufficient to train the deep learning networks for this application which forces to use of transfer learning [11]. Usually, transfer learning performs great but sometimes does not give satisfactory results when training data is not enough for fine-tuning those pre-trained deep learning networks. The focus of this work rests upon developing face recognition accuracy within different types of masks.

This writing is arranged as follows: Section 1 shows our face detection methodology. Section 2 describes the about-face embedding process. Section 3 narrates the classification procedure. Section 4 signifies the dataset. Section 5 presents test results analysis and section 6 draws concludes.


In this work, the FaceNet pre-trained model has been used for improving masked face recognition. We have benchmarked this approach with two well-known datasets and our dataset. Our approach tested on these datasets shows better recognition rates. So FaceNet model trained on masked and non-masked images gives better accuracy for simple masked face recognition. Although we concentrated on masks induced by a hat, sunglasses, beard, long hairs, mustache, and medical mask, our methodology can still be extended to more complex and many other sources of occlusion. Obviously, this method may not be appeasement for all types of masks. Further, a more accurate and sophisticated approach may than be needed. In later work, it is our importance to enhance and enlarge our work to address different extreme masks condition of face recognition.

At the same time Industry, 4.0, and/or Sustainable Technology will try to enhance the computer adoption and mechanization with autonomous and intelligent systems fed by data and machine learning. Working with data, security is essential. So our work can help these smart and autonomous industries to be more self-governing, secure, accurate, and efficient which helps more production and less waste.

About KSRA

The Kavian Scientific Research Association (KSRA) is a non-profit research organization to provide research / educational services in December 2013. The members of the community had formed a virtual group on the Viber social network. The core of the Kavian Scientific Association was formed with these members as founders. These individuals, led by Professor Siavosh Kaviani, decided to launch a scientific / research association with an emphasis on education.

KSRA research association, as a non-profit research firm, is committed to providing research services in the field of knowledge. The main beneficiaries of this association are public or private knowledge-based companies, students, researchers, researchers, professors, universities, and industrial and semi-industrial centers around the world.

Our main services Based on Education for all Spectrum people in the world. We want to make an integration between researches and educations. We believe education is the main right of Human beings. So our services should be concentrated on inclusive education.

The KSRA team partners with local under-served communities around the world to improve the access to and quality of knowledge based on education, amplify and augment learning programs where they exist, and create new opportunities for e-learning where traditional education systems are lacking or non-existent.

FULL Paper PDF file:

Masked Face Recognition Using Convolutional Neural Network



M. S. Ejaz and M. R. Islam,




Masked Face Recognition Using Convolutional Neural Network

Publish in

2019 International Conference on Sustainable Technologies for Industry 4.0 (STI), Dhaka, Bangladesh, 2019, pp. 1-6,



PDF reference and original file: Click here


+ posts

Somayeh Nosrati was born in 1982 in Tehran. She holds a Master's degree in artificial intelligence from Khatam University of Tehran.

Website | + posts

Professor Siavosh Kaviani was born in 1961 in Tehran. He had a professorship. He holds a Ph.D. in Software Engineering from the QL University of Software Development Methodology and an honorary Ph.D. from the University of Chelsea.

Website | + posts

Nasim Gazerani was born in 1983 in Arak. She holds a Master's degree in Software Engineering from UM University of Malaysia.