Face Recognition For Ensuring Security In University Campus
Face recognition, the technique to determine the individuals using facial images, not only a favored topic for research but also has plentiful application in the area of biometric, access control, information security, law enforcement, smart card, surveillance system etc. Bounteous favorable outcome has seen on special dataset through different model along with machine learning approaches. In paper, our aim was to apply Histogram of Oriented Gradient (HOG) on our dataset for live streaming on surveillance system applicable for security purposes to detect known and unknown face in real time data, and for this purpose we found 86.96% accuracy rate. Keywords—Face recognition, Dataset, HOG, Machine Learning.
Human has an extra ordinary power to identify a person in different expression, condition, in light variation. Now an artificial system is developed which will work like human known as Face Recognition and still now many work are processing in this field for finding better performance. Face recognition system mainly work for identifying faces by matching it with facial datasets. Face recognition system can be used mainly in two ways. Determine one and his details using his images from a large dataset of facial images. Here one‟s information is stored in a database with his images. One can find out him and his information by searching him with his images. Real time identification. While one passes though a surveillance system coverage area his basic information like name will show in the security monitor. This is varying in different condition. Face recognition has become a topic of interest in research since 1961. But in last two to three decades here looks a huge momentum in this field. Principal Component Analysis (PCA) was applied by Alex Pentland and Mathew Turk in 1991 to face classification .
A paradigm alteration has done by introducing Histogram of Oriented Gradient (HOG) by Navneet Dalal and B. Triggs . For person detection, HOG gives very good result as its dense overlapping grid. The advantage of HOG is fine orientation binning, fine scale gradient, relatively course spatial binning, well local contrast normalization which are important for good performance. Freeman and Roth used orientation histogram for hand gesture recognition . In the era of Neural Network more specifically by applying Deep Neural Network (DNN) and Convolutional Neural Network (CNN) this field finds a tremendous swiftness and sequel.
SHU Chang, and et al. used Histogram of Oriented Gradient achieving almost the same recognition rate with much lower computational time whereas this feature shows that single angle representation performs much better than double angle representation, applied this model on FRGC and CAS-PEAL databases . However still now many models are being introduced that‟s are applied in different dataset and sometime it is applied in some built in dataset for verifying their model and accuracy. The reminder of this paper is as follows. Section II presented the literature work that was highlighted our paper work. Section III for describing the methodology started by introducing the technology are used then the most important part dataset creation, uses the libraries in training and classification and at last result analysis. In section IV we discussed about challenges that we faced. Future works and conclusion is finally stated in section V.
In recent years many model has introduced for face recognition and image based work. Among them they show a huge success in this field. For this paper work we took knowledge from the following paper works. Taha J. Alhindi et al.  conducted a recent study which compares Local Binary Pattern (LBP), HOG and deep features from VGG 19, a pre trained deep network for feature extraction and Support Vector Machine (SVM), Decision Tree and Artificial Neural Network for the classification of histopathology image dataset, KIMIA Path960. The classification accuracy obtained in the study with LBP and SVM is 90.52%.
Hilton Bristow and Simon Lucey compared that the classifier which preserves local quadratic pixel interaction perfectly distinguishes between natural and noise than pixel based classifier . In this study a marvelous performance has seen with HOG and SVM having accuracy 99.3% applied in the Cohn Kanade + expression recognition dataset. Kanade presents Automatic feature extraction using ratio of distances gained accuracy rate 45%-75% with a dataset of 20 people . A set of geometrical feature was computed by Brunelli, Roberto and T. Poggio like mouth position, chin length, nose width and length . They had the accuracy rate 90% on a dataset of 47 people.
Wiskkot and et al. compared 300 faces against 300 different faces of the same people taken from the Face Recognition Technology (FERET) dataset not only for face recognition but also for identifying gender. They record a recognition rate of 97.3% . For matching two image of same person in unconstrained environment, (Convolutional neural network) Covnet-Restricted Boltzmann Machine has shown 97.08% accuracy . A mixture distance technique achieved recognition rate 95% using a query dataset of 95 images (each image has 30 extracted feature) introduced by Cox et al. . Hand crafted features such as Local Binary Pattern (LBP) has received a respectable performance in Face Recognition in constrained environment . But the performance is deteriorated when the images are taken in unconstrained environment.
Harihara Santosh Dadi and et al. introduce an improved facial recognition rate using HOG feature and SVM classifier having an improvement of 8.75% face recognition rate. They applied the formula on color FERET, Yale Database, Yale face Database „B‟, BioID, Georgia Tech, FEI, leveled faces in the wild and get 68.5, 98.2, 88.6, 75.67, 81.25, 80.13, 64.6 percent accuracy respectively . In this proposed paper, our goal was to experiment on our own created dataset using OpenCV, Dlib, Face Recognition library and HOG classifier for feature extraction and classification in real time data for recognizing intruders.
A facial recognition system is a computer application capable of identifying or verifying a person from a digital image or a video frame from a video source. One of the ways to do this is by comparing selected facial features from the image and a face database. The proposed system is supposed for real time live streaming surveillance system to recognize the individuals and also detect known and unknown for informing the doer of the team related with security purposes which will differ from the conventional system. The different implementation methods regarding our system are described in details in below sections.
Different technologies are being used to make the system widely available through cross-platforms and to provide best performance within limited resources. These help to build system rapidly and help to maintain consistence that helps developers to meet their deadlines. Among the technology used to build the system, some of them are explained in the next sections.
OpenCV known as Open Source Computer Vision is a leading open source library for computer vision, image processing and video processing. It also supports machine learning which enables this tool to be used for object detection from videos or images. It can also input/output videos in real time. OpenCV provides different machine learning algorithms for face recognition from images. For the wide range of utilities, we are using OpenCV interface for Python to process face identification in real time. Python: In this proposed system Python is used for building our system. Python is an open source programming language. Python is easy-to-read and powerful. Its high-level built in data structures, combined with dynamic typing and dynamic binding, make it very attractive for Rapid Application Development. For its beauty, simplicity, power and availability of powerful modules and packages for modern technology like Data Analysis, Artificial Intelligence, IoT and many more, we are building our system on Python.
Dlib is a library which is a C++ toolkit containing machine learning algorithm and tools for creating complex software to solve real world problem. Wide range of domains including embedded system, robotics, mobile phone and large high performance computing environments it is used. It is used to construct our face embeddings used for the actual recognition process. To train a network from scratch, huge data will be needed but it is easier to use a pre trained network and then use it to construct 128-d embeddings for each faces in dataset. Here the network quantifies faces constructing 128-d embedding for each. Face Recognition: Face recognition is a library which wraps around dlib‟s facial recognition functionality. Dlib has given the facility to use variety of machine learning algorithm but in face recognition library in dlib allow two model for facial recognition that‟s are Convolutional Neural Network (CNN) and Histogram of Oriented Gradient (HOG).
CONCLUSION AND FUTURE WORK
In this paper, we implemented a basic real time face recognition system. This system can be further improvised to create more complex and advanced system for reaching any particular goal. There are many areas of improvements for this project. With APIs like Tensorflow, there are more powerful algorithms that use Deep Neural Networks for processing data, providing more accuracy over the given data. By improving these areas, we can build our system to be more powerful and ready to be used in any existing system. In future, we must need to focus on the challenges. In the proposed system, HOG is used but using CNN more perfect result will be found. This system can play a vital rule in security purposes everywhere in our country. In traffic control this can also give a great support. In the classroom it can be used to take attendance so the time can be saved and proxy will not possible anymore.
- M. Turk and A. Pentland, “Face Recogniion Using Eigenfaces,” Proceedings of CVPR IEEE Computer Society, Jun. 1991, pp. 586-591.
- Navneet Dalal and Bill Triggs, “Histogram of Oriented Gradients for Human Detection,” IEEE Computer Society Conference on Computer Vision and Pattern Recognition, Jun 2005, pp. 886-893.
- Freeman W T, Roth M. Orientation histograms for handgesture recognition. In: Intl. Workshop on Automatic Face and Gesture Recognition. IEEE Computer Society, Zurich, Switzerland, 1995: 296-301.
- SHU Chang, DING Xioqing, FANG Chi, “Histogram of the Oriented Gradient for Face Recognition*,” TSINGHUA SCIENCE AND TECHNOLOGY ISSN 1007-0214 15/15 pp 216-224, vol. 16, Number 2 April 2002.
- Taha J. Alhndi, Shivam Kalra, Ka Hin Ng, Anika Afrin, Hamd R. Tizhoosh, “Comparing LBP, HOG and Deep Features for Classificaton of Histopathology Images,” accepted for publcation in proceedings of the IEEE World Congress on Computational intelligence (IEEE WCCI), Rio de Janeiro, Brazil, July 2018.
- Hilton Bristow, Simon Lucey (2014), “Why do linear SVMs trained on HOG features perform so well?”
- Kanade, Takeo. “Picture processing system by computer complex and recognition of human faces” Doctoral dissertation, Kyoto University 3952 (1973): 83-97.
- Brunelli, Roberto, and Tomaso Poggio. “Face recognition: Features versus templates.” IEEE transactions on pattern analysis and machine intelligence 15.10 (1993): 1042-1052.
- Laurenz Wiskkot, Jean Marc Fellous, Norbert kruger, Christoph von der Malsburg, “Face recognition and gender determination*,” Proceedings of the Intern. Workshop on Automatic Face- and Gesture-Recognition, 1995, Zurich, pp. 92-97.
- Srivastava, Nitish, et al. “Dropout: a simple way to prevent neural networks from overfitting.” Journal of Machine Learning Research 15.1 (2014): 1929-1958.
- Cox, Ingemar J., Joumana Ghosn, and Peter N. Yianilos. “Feature-based face recognition using mixture-distance.” Computer Vision and Pattern Recognition, 1996. Proceedings CVPR’96, 1996 IEEE Computer Society Conference on. IEEE, 1996.
- Ahonen, Timo, Abdenour Hadid, and Matti Pietikäinen. “Face recognition with local binary patterns.” Computer vision-eccv 2004 (2004): 469-481.
- Harihara Santosh Dadi, Gopala Krishna Mohan Pillutla, “Improved Face Recognition Rate Using HOG Features and SVM Classifier,” IOSR Journal of Electronics and Communication Engineering (IOSR-JECE) e-ISSN: 2278-2834,p- ISSN: 2278-8735.Volume 11, Issue 4, Ver. I (Jul.-Aug .2016), PP 34-44.
Cite this Essay
To export a reference to this article please select a referencing style below