Content uploaded by Shubha P Jagi
Author content
All content in this area was uploaded by Shubha P Jagi on Oct 15, 2020
Content may be subject to copyright.
Human Face Recognition using Local Binary Pattern algorithm - Real time
Validation
1Shubha. P,2 Dr. M. Meenakshi
1,2 Dept. of Instrumentation Technology,
1,2 Dr.Ambedkar Institute Of Technology,
1,2 Banglaore-560056, India,
1shubhajagi@gmail.com
2meenakshi_mbhat@yahoo.com
Abstract: A real time face recognition using LBP algorithm and image processing techniques are proposed. Face image is
represented by utilizing information about shape and texture. In order to represent the face effectively, area of the face is split
into minute sections, then histograms of Local Binary Pattern (LBP) are extorted which are then united into a single histogram.
Secondly, the recognition is carried out on computed feature space using nearest neighbor classifier. The developed algorithm is
validated in real time by developing a prototype model using Raspberry Pi single board computer and also in simulation mode
using MATLAB software. The above obtained results match with each other. On comparing both the results, recognition time
taken by the prototype model is more than that of the simulation results because of hardware limitations. The real time
experimental results demonstrated that the face recognition rate of LBP algorithm is 89%.
Keywords: Local Binary Pattern (LBP), MATLAB, histogram, classifier, rRaspberry Pi, Real time face recognition.
I.INTRODUCTION:
Face recognition is a popular image processing application. Using this method a person can be identified and
verified with the database. The process identification involves face recognition, which compares unknown face
image with the large set of face image database. The obtained value is the measure of the similarity between two
images. The face image is recognized when the obtained value is equal to or greater than the threshold value. The
recognition process includes the following three stages. Firstly, Face representation carries out the process of
modeling a face. Secondly, feature extraction stage includes extraction of unique and useful features of a face.
Lastly, classification stage compares the face image with database images.
Applications of face recognition includes in various fields such as banking, airports, passport office, government
offices, security systems and many more. There are two phases of face recognition. The first phase includes training
of the faces which are saved in the database and second phase is the verification stage to obtain the exact image of
the face.
Face recognition technique is considered as a challenging method due to variation in size, color, pose variations,
illumination variations, and face rotation. To overcome the above problem Local Binary Pattern algorithm (LBP) is
proposed. The paper is organized in the following steps: First, Existing methods of face recognition is given in
section II. Next, in section III and IV the proposed method and face recognition algorithm are explained. Section
V and VI gives the working principle and the results of the experiment. Finally section VII is referred in order to
draw conclusions.
II. EXISTING TECHNIQUES OF FACE RECOGNITION:
The various methods are available for face recognition which includes: Principal Component Analysis (PCA) and
Linear Discriminant Analysis (LDA).
In PCA algorithm [2] the number of variables can be reduced statistically. The relevant features are extracted from
the face. In the training set, the obtained images are combined as weighted eigen-vectors and eigenfaces. The
covariance matrix of images in training database provides the eigen-vectors known as basic function and the
corresponding weight is determined by choosing a set of relevant eigenfaces. Values of eigen are increased to
obtain higher face recognition.
Classification of images according to a group of features is carried out by LDA[2,3]. Between class scatter matrix
(SB) and within class scatter matrix (SW) of all the classes of all samples are defined. The differences between
classes are found for recognition. The above methods cannot produce complete information of face due to the loss of
information and they are sensitive to noisy image and lighting conditions in the images [4]. PCA algorithm fails to
recognize the identical twin faces.
III. PROPOSED METHOD
In order to overcome the above problems, local binary pattern algorithm is proposed [5]. The algorithm is less
sensitive to illumination variations and scaling variations. LBP are very efficient approach for feature extraction. It
encodes differences between pixel intensities in a local neighborhood of a pixel. Tiny patterns from a face image are
identified by applying the local binary pattern operator [6].
Face recognition system includes the following modules. Firstly, image acquisition module in which the details of
the image are gathered and these samples are taken for analysis. Secondly, feature extraction module extracts only
significant features from the image. Lastly, classification and training classifier database module classifies the
obtained samples.
i.Face detection:
Face detection is represented by a block diagram as given in Fig.1. A face can be detected by using Haar cascade
classifier from OpenCV. Detection of various organs on the face is carried out by Haar cascade classifier which
operates based on AdaBoost algorithm. The detected faces are stored in a FR100.
Fig.1: Block diagram representing detection of a face
ii. Face Recognition:
The technique of face recognition is depicted in Fig. 2. The calculation of the binary pattern of a face is carried out
by converting it into a gray scale image[7].The LBP operator is not fixed with the radius and neighborhood.
Therefore the algorithm can be applied to the images of different size and texture. The face description is
represented through feature vector. The histograms for the region are calculated. The united histogram represents an
overall description of the face [8].
Fig.2: The block diagram representation of the face Recognition technique.
iii. Database:
Database FR100 is created to recognize a human face. The database stores 100 images of each face with different
facial expressions and postures in an image. During acquisition process database images are converted to gray scale
images for feature extraction. Then the images are normalized for better recognition results. Each image undergoes a
process of normalization to eradicate noise and to align it in proper position.
IV. BLOCK DIAGRAM & WORKING PRINCIPLE OF HARDWARE SETUP
Fig.3: Block diagram representation of Hardware setup
Fig. 4: Working principle for recognition of a face
The hardware setup is shown in Fig. 3 whereas working principle of the face recognition system is depicted by
means of a block diagram as shown in Fig. 4. The hardware module consists of Raspberry pi single board computer
powered by a micro USB socket with 5 Volts, 2.5Amps.The LBP algorithm is made to run in python IDE. The
image is captured by a camera. The obtained color image is converted into grayscale image for further processes.
The required facial region is cropped and is divided into several blocks. The unique ID is assigned to each image in
the database as a result a database is created. Then the extraction of texture features of LBP from the test image is
carried out. A face is recognized by using Harr cascade classifier and recognizer. Database image is used by the
classifier to compare it with the test image. Face recognition is successful only when the test image matches with the
images of the database.
V.FACE RECOGNITION ALGORITHM
Local binary algorithm is applied to perform face recognition. Proposed algorithm is as described below:
1. Inputting of the face image.
2. Dividing the face image into blocks
3. Calculating histogram for each block
4. Combining LBPH histogram into single histogram
5. Comparing the face image with the database image for further processing
6. Successful recognition of the face only when the test image matches with the images of the database.
VI.RESULTS & DISCUSSIONS
Algorithm is tested for its performance using different test images in simulation environment and in hardware setup.
Before applying an algorithm, required face region is cropped as depicted in Fig. 5(a). Input image and LBP Image
are illustrated in Fig. 5(b). The recognition result is represented in Fig. 5(c). The hardware setup result is described
in Fig. 6.
Fig.5: Simulation results of LBP algorithm (a) Cropped Image, (b) Original Image and LBP Image, (c) Detected Image
Fig. 6: Face Detected from experimental Setup
The overall rate of recognition of an LBP algorithm is 89% which is observed in Table I. The experiment is also
extended for images with varying positions of the face. From table II, it can be concluded that rate of recognition of
a face facing the front is greater as compared to the other positions of the face.
Table I: Rate of recognition of FR100 database
No. of
Images in
the
database
No. of
input
faces
tested
Recogni
zed face
Un
Recogniz
ed face
Recogni
tion
Rate
5500
55
49
06
89
TABLE II: Recognition rate of FR100 database at various position of face
Left
Front
Right
58
89
45
The results of table III depicts that the recognition rate is high in LBP algorithm when compared to the other face
recognition algorithms such as PCA, 2D-PCA and LDA. This is represented in Fig.7.
Table III: Face Recognition rate of standard face recognition algorithm [9]
Standard face
recognition methods
Recognition
Rate (%)
LBP
89
PCA
64
2D-PCA
63.1
LDA
55
Fig.7: Comparison of various algorithms
VII. CONCLUSION
The paper presented a simple and efficient way of face recognition using LBP algorithm. The design and validation
of real time face recognition is proposed. The proposed system is developed by using LBP algorithm and techniques
of image processing. MATLAB software is used to validate the results in offline mode whereas experimental setup
is used to validate the results in real time. These results match with each other. The rate of recognition of a face by
LBP algorithm is obtained as 89% by which we can prove that above algorithm is better when compared to other
face recognition algorithms. The same work can be extended to an autonomous robot.
REFERENCES
[1] Sarabjit Singh, Amritpal Kaur, Taqdir, "FaceRecognition Technique using Local Binary Pattern Method",
International Journal of Advanced Research in Computer and Communication Engineering Vol. 4, Issue 3, March
2015.
[2] Shubha P, M.Meenakshi, presented a paper entitled “Design and development of image processing algorithms-A
comparison” in an 1stInternational conferences on Advances in Science & Technology ICAST-2018 in association
with IET , India, April 6th-7th2018,Mumbai.,ISSN 0973-6107.
[3]Riyanka, “Face Recognition Based on Principal Component Analysis and Linear Discriminant Analysis”,
IJAREEIE, vol.4, Issue 8, August 2015
[4]Gonzalez and Woods, “Digital ImageProcessing”,3rd Edition, Prentice Hall, Pearson Education, USA ,2008.
[5] Roberto Bruunelli and Tomaso Poggio, "Face Recognition: Features versus Templates", IEEE Transactions on
Pattern analysis and Machine intelligence Vol.15.No.10, October 1993.
[6] P.B.Khanale, Recognition of Marathi Numerals Using Artificial Neural Network, Journal of Artificial
Intelligence 3(3): 135-140, 2010.
[7] A. Rahim, N. Hossain, T. Wahid and S. Azam, "Face Recognition using Local Binary Patterns," Global Journal
of Computer Science and Technology Graphics & Vision, vol. 13, no. 4, pp. 1-9, 2013.
[8] D. Huang, C. Shan, M. Ardabilian, Y. Wang and L. Chen, "Local Binary Patterns and Its Application to Facial
Image Analysis: A Survey," IEEE Transactions on Systems, MAN and Cybernetics-Part C: Applications and
Reviews, vol. 41, no. 6, pp. 765-781, November, 2011.
[9]T S Vishnu Priya, G.Vinitha Sanchez, N.R.Raajan, ”Facial Recognition System Using Local Binary
Patterns(LBP)”, International Journal of Pure and Applied Mathematics, Volume 119 No. 15 2018, 1895-1899