MSB based Face Recognition Using Compression and Dual Matching Techniques

Vasantha Kumara M & Mohammed Rafi
International Journal of Image Processing (IJIP), Volume (13) : Issue (4) : 2019 62
MSB based Face Recognition Using Compression and Dual
Matching Techniques
Vasantha Kumara M cmn.vasanth@gmail.com
Assistant Professor/Dept. of CSE
Govt. SKSJTI
Bangalore-01, India
Mohammed Rafi mdrafi2km@yahoo.com
Professor/Dept. of Studies/CSE
University BDT College of Engineering
Davangere-572004, India
Abstract
Biometrics are used in almost all communication technology applications for secure recognition. In
this paper, we propose MSB based face recognition using compression and dual matching
techniques. The standard available face images are considered to test the proposed method. The
novel concept of considering only four Most Significant Bits (MSB) of each pixel on image is
introduced to reduce the total number of bits to half of an image for high speed computation and less
architectural complexity. The Discrete Wavelet Transform (DWT) is applied to an image with only
MSB's, and consider only LL band coefficients as final features. The features of the database and test
images are compared using Euclidian Distance (ED) an Artificial Neural Network (ANN) to test the
performance of the pot method. It is observed that, the performance of the proposed method is better
than the existing methods.
Keywords: Biometrics, Face Recognition, DWT, MSB, Compression.
1. INTRODUCTION
The face recognition machinery has been extensively used in every phase of social life such as
forensic applications, human-computer interaction, bank transactions, land records verification,
password serviced etc. The main challenging task in face recognition system is to identify human
beings with variations in face images due to illumination changes, angle variations in face images,
shape variations due to age and environmental changes etc. The performance of the face recognition
system depends on feature extraction techniques and classification techniques. There are mainly
three types of features such as spatial domain features, transform domain features and hybrid domain
features. In spatial domain feature extraction approach, the pixel values are manipulated directly
using techniques such as standard deviation, variance, Principal Component Analysis (PCA) [1],
Local Binary Pattern (LBP) [2] etc. In transform domain feature extraction, the spatial domain is
converted into the frequency domain and the corresponding frequency coefficients are manipulated to
derive transform domain features. The examples of frequency domain techniques are Fast Fourier
Transform (FFT) [3], Discrete Cosine Transform (DCT) [4], Discrete Wavelet Transform (DWT) [5] etc.
The spatial domain and transform domain features are fused to obtain hybrid coefficients in hybrid
domain techniques [6].
Contribution: In this paper, MSB based face recognition using compression and dual matching
technique is proposed. Only four bits of each pixel are considered for feature extraction using DWT.
The dual matching technique such as ED and ANN are used to compare performance parameters.
Organization: The rest of the sections are as follows, chapter 2 contains the literature survey, chapter
3 gives the description of our proposed model, chapter 4 provides the algorithm for the proposed
model. In Chapter 5 we evaluate the performance parameters and also compared proposed method
with the existing methods, chapter 6 enunciates the conclusion.

2. LITERATURE SURVEY
In this section, the literature appraisal of current techniques to distinguish the human beings based on
physiological biometric trait face pictures are explained. The numerous methods of preprocessing,
feature extraction and matching presented by many investigators are given.
Vikash, and Rajesh Parihar [7] presented evaluation between DFT and Wavelet based image
alteration methods. It is based on approximation and elimination of distortion in the altered images.
The comparison was done between DFT based JPEG and Wavelets based JPEG 2000 modification
techniques on the Image Quality, Modification Ratio, MSE and PSNR etc. The, distortion cannot be
avoided during visual data acquisition, processing and transmission and exhibit as the random
variation of brightness or color in images and should be removed for image quality assessment,
restoration and enhancement. Thamizharasi Ayyavoo and Jayasudha John Suseela [8] presented
Discrete wavelet transform enhanced contrast limited adaptive histogram equalization (DWT E-
CLAHE) in the illumination pre-processing to identify the front view face images. The face images are
upgraded using the Gamma intensity correction (GIC) and split into low-frequency and high-frequency
components using 2D DWT. The logarithmic transform, GIC and CLAHE are applied in the sequential
order low-frequency components. The face recognition of DWT ECLAHE is made using Gabor
magnitude features.
Zied Bannour Lahaw et al., [9] proposed face recognition using PCA, ICA, LDA based on DWT, and
SVM algorithms. The DWT is used in preprocessing to compress input images and considered
approximation band image for further processing. The PCA, ICA, and LDA are used to extract
features. The classifier SVM is used to categorize the images for recognition. Govardhan Mattela and
Sandeep K. Gupta [10] presented Gabor-mean-DWT for automatic facial expression recognition from
video image sequences and its illumination invariant. Gabor filter is used to extract edges and texture
pattern features of face images, but the limitations of huge dimension and high redundancy. This
limitation is reduced by average-DWT feature reduction technique. Jyothi Ravikumar et al., [11]
proposed the convolution based feature extraction technique of face recognition using Discrete
Wavelet Transform (DWT) and Histogram of Oriented Gradient (HOG). The DWT is applied on face
images to a obtained compressed number of transform domain LL band coefficients. The HOG is
used on the LL band to further compress LL coefficients to obtain oriented gradients. The ultimate
features are obtained by convolving LL and HOG coefficients. Sumathi and Christopher Derairaj [12]
proposed the occlusion, pose and illumination variations face recognition. The Principal Component
Analysis and DWT are applied to extract the features. The SVM is used for classification and on the
way to compute performance parameters of the system. Rifha Ilyas Bendjillali et al., [13] proposed the
facial expression and emotion recognition. The method applied Viola-Jones as face detection,
Contrast Limited Adaptive Histogram Equalization to adjust the histogram and obtained the extracted
features using DWT. The classification stage is applied Convolution Neural Network as the classifier.
Taif Alobaidi and Wasfy B. Mikhael [14] proposed a face recognition system based on two domains.
The DWT is applied on face images to obtain initial features and applied DCT to achieve the final
features. The ED matching technique is used to calculate the minimum matching distance among
training features and testing features.
3. PROPOSED METHOD
In this section, the novel approach of binary bit segmentation of pixels and considered Most
Significant Bits (MSB) for further processing to extract features is proposed. The method includes pre-
processing, feature extraction and classification. The proposed approach of face identification system
is shown in Figure 1.

Concatenation
Zero
4 Bits
Database
Binary Bit Segmentation
MSB
4 Bits
LSB
4 Bits
DWT
LL LH HL HH
LL
MSB 8 Bits
Matching
Concatenation
Zero
4 Bits
Test
Binary Bit Segmentation
MSB
4 Bits
LSB
4 Bits
DWT
LL LH HL HH
LL
MSB 8 Bits
Result
FIGURE 1: Proposed Approach of Face Recognition.
3.1 Face Database
The performance of the proposed face recognition model is evaluated using various typical face
databases such as ORL, YALE, EYB, and JAFFE.

3.1.1ORL Database [15]: It is a typical face database contains face images of ten dissimilar images
of each of 40 distinct persons ie., 400 images in total. The images are taken under changeable
lighting, dissimilar facial expressions and various facial particulars. All the images of an individual are
captured in an upright and frontal position. The size of each image is of 92x112 pixels. The ORL face
image samples of one person is shown in Figure 2.
FIGURE 2: ORL Face Database Samples of One Person.
3.1.2 Japanese Female Face Expression (JAFFE)[16]: In the Japanese Female Face Expression
(abbreviated as JAFFE), database there are twenty images of each of 10 individual subjects, 200
images in total. The size of each image is 256 x 256 pixels, the 2D nature of the matrix. All images
were taken against an upright, Frontal position. The JAFFE face image samples of one person is
shown in Figure 3.
FIGURE 3: Sample Face of JAFFE Face Database.
3.1.3 Yale [17]: The face database consists of 165 face images of 15 subjects. Each individual is
having different facial expression and captured with occlusion. Image resolution of this database is
320×243.
FIGURE 4: Sample Face of YALE Face Database.

3.1.4 EYB [18]: The Extended Yale B (EYB) face database contains 2141 face images of 38 subjects.
The face images are captured under altered illumination circumstances. All the images are in cropped
version and resize to the size of 192X168.
FIGURE 5: Sample Face Images of the EYB Face Database.
3.2 Binary Bit Segmentation of Pixels
The novel concept of effective MSB decimal values utilization of pixels to reduce 256 gray levels to 16
gray levels is introduced in the proposed model for effective face recognition. The images are
collected from face databases and the decimal values of each pixel are converted into binary of eight
bits. The binary bits of every pixel are segmented into two parts ie., four bits Least Significant Bits
(LSB) and four bits Most Significant Bits (MSB). The four MSB bits of every pixel contain significant
information of an image while the four LSB bits having insignificant information of an image. The MSB
bits only considered for feature extraction by replacing four LSB bits with zeros to convert four MSB
bits to eight bits. The decimal equivalent values of four-bit MSB varies between 0 and 240 with only
sixteen gray scale values using (1). By considering four MSB bits, the technique rejects 240 gray
levels and retain only 16 gray levels. The decimal values of four MSB bits are significant compared to
original pixel values of an image.
(1)
Where n is the bit position from the right side.
(a) Original Image (b) MSB Image (c) LSB Image
FIGURE 6: Binary Segmentation Images.
The binary segmentation of an image is shown in Figure 6, it is seen that the image corresponding to
MSB is almost same as the original image, whereas the image corresponding to LSB not same as the
original image. The lead of binary segmentation is to increase the speed of computation and also
decrease complexity is an architecture for real time systems.

3.2.1 Discrete Wavelet Transforms (DWT)
The transformation is employed on the rows of the image using a low pass filter and high pass filters
concurrently. The detailed part (high frequency) and approximation part (low frequency) are obtained
through the filters and down sampled by a factor of 2. The operation is repeated with the columns of
an image. The four sub-images obtained as an approximation (LL), vertical (LH), horizontal (HL) and
diagonal (HH) bands for every level of decomposition. The wavelet coefficients are used in three ways
as (i) All the four band coefficients are used directly as features. (ii) Only approximation band
coefficients are used as features. (iii) The approximation band coefficients are combined with other
types of features to obtain final features. The horizontal subbands are insensitive to the varying
illuminations while approximation subbands are insensitive to the varying facial expressions [19].
Sellahewa and Jassim [20] had shown that the approximation subband is suitable for recognition
under controlled illumination, but it is affected by the varying illuminations. The horizontal and vertical
subbands are reasonably robust to the varying light. However, these were affected by geometrical
changes like varying facial expressions and pose. The identification of facial images under poor and
uncontrolled lighting is a challenge for horizontal and vertical subbands. The DWT is applied to MSB
image and the corresponding four band images are shown in Figure 7. It is realized that, the LL band
is almost same as the original image and consists of significant information. The images
corresponding to LH, HL, and HH are not same as the original image and has insignificant
information. The features for classification of face images are extracted from the LL band only, which
reduces the number of features.
(a) MSB Image
(b) LL Image (c) LH Image (d) HL Image (e) HH Image
FIGURE 7: DWT on MSB Image.
3.2.2 Matching
In this paper, Artificial Neural Network (ANN) and Euclidean Distance (ED) are used to classify the face
images using the proposed algorithm.
1. ANN: It is computational models capable of machine learning as well as pattern recognition. It has
3 layers as input layer, hidden layer and output layer. The efficiency of the proposed model using
ANN is shown by plotting the confusion matrix, which is the simplest metric used to determine the
precision and accuracy of the model. There are four terms associated with confusion matrix to
compare the cases between the actual class and the predicted class as True Positives (TP), True
Negatives (TN), False Positive (FP) and False Negative (FN). In order to determine the
performance of the proposed model, there are two significant parameters.
i. Classification Accuracy: It is the proportion of the complete number of correct predictions
against number of training images is given in (5).
(5)

ii. Test Accuracy: It is the proportion of the complete number of correct predictions against
number of testing images is given in (6).
(6)
2. ED: There are four parameters used in ED Matching to test and analyze the performance of the
proposed model are as follows
i. False Acceptance Rate (FAR): It is the ratio of number of Persons Outside Database
(POD) falsely accepted as Persons Inside Database (PID) to the total number of POD as
given in Equation (7).
(7)
ii. False Rejection Rate (FRR): It is the ratio of number of Persons Inside Database (PID)
falsely rejected as Persons Outside Database (POD) to the total number of PID as given in
Equation (8).
(8)
iii. Equal Error Rate (EER): It is the value where both FAR is equal to FRR. The values of
FAR and FRR are equal at optimum threshold value.
iv. Total Success Rate (TSR): It is the ratio of the total number of PID are correctly matched
to the total number of PID. The TSR can be computed using Equation (9).
(9)
v. Optimum Total Success Rate (OTSR): The value of TSR corresponding to EER.
Vi. Maximum Total Success Rate (MTSR): The maximum value of TSR.
4. PROPOSED ALGORITHM
Problem Statement: The robust algorithm for face recognition using only four MSB’s of image pixels
for feature extraction is developed in identifying a person.
The proposed algorithm is given in Table-1.
Objectives: Face recognition algorithm is developed with the following objectives
i. To reduce FRR, FAR and EER
ii. To increase TSR
iii. To increase speed of computation and reduce hardware complexity by compressing
number of features.
Input: Standard Face Images
Output: Performance parameters
1 Various face image databases viz., ORL, YALE, EYB, JAFFE and L-SpaceK are used
to test the algorithm
2 The eight bits of pixels are segmented into four MSB and four LSB groups
3 Only four MSB’s are considered for feature extraction, which reduces biometric system
complexity and also increases speed of computation
4 The DWT is applied on image matrix with only four MSB’s for each pixel
5 The LL sub-band coefficients of DWT is considered as features
6 The final features are compressed version with only MSB’s and DWT
7 The database image features and test image features are compared using ED and
ANN to compute performance parameters
8. The computer performance parameters of proposed method are better than the
existing methods.
TABLE 1: Proposed Algorithm.

5. EXPERIMENTAL RESULT
In this section, the performance evaluation of the proposed method is tested using ORL, YALE, EYB,
and JAFFE face databases.
5.1 Performance Analysis of Proposed Model using The ED Matching Technique
The performance parameters such as EER, OTSR and MTSR using ORL, Yale, EYB and JAFFE face
databases are computed. The total Number of Persons (NOP), Images Per Person (IPP) and size of
each face image for face databases are given in Table 2. The performance parameters for various
combinations of Persons Inside Database (PID) and Persons Outside Database (POD) for face
databases are tabulated. The final number of features used to recognize face images are 1/4
th
of
initial dimensions of original face images. It is observed that, the percentage MTSR for all four face
databases are maximum for almost all combinations of PID and POD’s. The percentage OTSR is
maximum in the case of EYB face database, whereas lowest in the case of JAFFE face database.
The percentage EER is higher in the case of JAFFE and lower in the case of EYB face database. The
percentage OTSR decreases with increase in PID. The percentage EER increases with increase in
PID and decrease with POD.
Databases NOP IPP
Image
size
No. of
Feature
IPP
(Train)
PID POD
Optimum
Threshold
EER OTSR MTSR
ORL 40 10 112X92 2576 8 10 30 0.32 10 90 100
(10304) 10 20 0.32 10 90 100
10 10 0.364 10 90 100
20 10 0.319 15 85 100
30 10 0.292 16.67 83.33 100
Yale 15 10 243X320 19520 9
(77760)
7 8 0.1259 12.5 85.71 100
8 7 0.1259 14.29 75 87.5
10 5 0.1258 20 80 90
EYB 31 64 192X168 8064 62
(32256)
15 11 0.788 6.67 93.33 100
15 15 0.747 13.33 86.67 100
20 11 0.7423 15 85 100
JAFFE 10 20 256X256 16384 19
(65536) 3 7 0.92 33.33 66.67 100
5 3 0.79 20 80 100
5 5 0.80 20 80 100
7 3 0.95 14.29 85.71 100
TABLE 2: Recognition Rate of proposed model using ED with various face databases.
The variations of FRR, FAR and TSR with threshold values for ORL, YALE, EYB and JAFFE face
databases are plotted in Figures 8, 9, 10 and 11. It is observed that as threshold value increases, the
values of FRR values decrease from maximum to minimum. The values of the FAR and TSR
increases with increase in threshold values. The value of the EER is 10% on the optimal threshold
value of a 0.364 for an ORL database with a PID and POD combination of 10 and 10. The value of
the EER is 12.5% on the optimal threshold value of 0.1259 for a YALE database with a PID and POD
combination of 7 and 8. The percentage values of EER and OTSR of 6.67 and 93.33 at optimum
threshold value of 0.788 for EYB database with PID and POD combinations of 15 and 11. At the
optimum threshold of 0.95, the corresponding percentage EER and OTSR values are 14.29 and
85.71 with PID and POD combination of 7 and 3 for JAFFE face database. The optimum threshold
values are lower in the case of YALE face database, whereas higher values in the case of JAFFE
face database.

FIGURE 8: Variations of performance parameters for PID &POD of 10&10 using ORL Database.
FIGURE 9: Variations of performance parameters for PID &POD of 7&8 using YALE Database.

FIGURE 10: Variations of performance parameters for PID &POD of 15&11 using EYB Database.
FIGURE 11: Variations of performance parameters for PID &POD of 7&3 using JAFFE Database.

5.2 Performance Analysis of Proposed Model using ANN Classifier
The overall accuracy of the proposed method is computed using ORL, Yale, EYB, JAFFE and L-
spaceK face databases. The total Number of Persons (NOP), Images Per Person (IPP) and size of
each face image for face databases are tabulated in Table 3. The overall accuracy is calculated by
varying images per person (IPP) in training by keeping the number of hidden layers constant for each
face database. The number of final features are to almost 1/4
th
of original image dimensions. It is
observed that, the overall accuracy increases with an increase in the number of images per person in
training. The performance of the proposed method is better using EYB face database.
Database NOP IPP Image size
No. of
Feature
Hidden
Layers
IPP
(Train)
Overall
Accuracy
ORL 40 10 112X92 2576 22 2 52.5000
(10304) 4 71.2500
5 80.2500
6 88.2500
8 98.2500
Yale 15 10 243X320 19520 19 2 72.6667
(77760) 4 91.3333
5 96.0000
6 99.3333
8 99.3333
EYB 31 64 192X168 8064 25 10 94.5060
(32256) 20 100.0000
32 100.0000
40 100.0000
50 100.0000
Jaffe 10 20 256X256 16384 23 4 77.5000
(65536) 8 95.5000
10 98.5000
12 99.5000
16 99.0000
L-spaceK 152 20 200X180 9000 25 4 51.6447
(36000) 8 79.1118
10 88.4539
12 94.5395
16 99.3092
TABLE 3: Overall Accuracy of proposed model using ANN for various face databases.
5.3 Performance Comparison of The Proposed Method using ED and ANN Matching
Techniques
The percentage recognition rate of the proposed method using ORL, Yale, EYB and JAFFE face
databases with ED and ANN matching techniques is given in Table 4. It is observed that, the ANN
matching technique is better compared to ED technique. The recognition rate is almost 100% in the
case of ANN matching technique for all face databases.
TABLE 4: Percentage Recognition Rate comparison of proposed method using ED and ANN.
5.3 Performance Comparison of Proposed Method with Existing Methods
The accuracy of the proposed method is compared with existing methods for face databases such as
ORL, YALE, JAFFE and EYB use ANN matching technique in Table 5. The percentage accuracy is
98.05 which is high compared to 91.5 and 95.25 of existing methods presented by W. Yu et al., [21]
Database
Percentage Recognition Rate
ED ANN
ORL 90 98.2500
Yale 85.71 99.3333
EYB 93.33 100.0000
Jaffe 85.71 99.50

and Sharif et al., [22] using ORL database. The proposed method with ANN is compared with existing
methods presented by Bhaskar Anand and Prashant K Shah [23] and Riddhi A. and S.M. Shah [24]
using YALE face database and observed that our method has better accuracy. The existing methods
presented by Biao Yang et al. [25] and Yuanyuan Ding et al. [26] are compared with our method using
JAFFE face database and observed that our method gives better accuracy of 99.50. The EYB face
database is used to compare the accuracy of the proposed method with existing methods presented
by Yuqi Pan and Mingyan Jiang [27] and Shih-Ming Huang and Jar-Ferr Yang [28] to show that our
method is superior and has accuracy 100%. In a summary, the proposed method is better compared
to existing methods for ORL, YALE, JAFFE and EYB face databases.
Database Author Accuracy (%)
ORL W. Yu et al. [21] 91.50
Sharif Muhammad et al. [22] 95.25
Proposed with ANN 98.25
YALE Bhaskar Anand and Prashant K Shah [23] 97.78
Riddhi and Shah [24] 98.18
JAFFE Biao Yang et al. [25] 92.21
Yuanyuan Ding et al. [26] 94.84
EYB Yuqi Pan and Mingyan Jiang [27] 91.60
Shih-Ming Huang and Jar-Ferr Yang [28] 96.37
Proposed with ANN 100
TABLE 5: Percentage Accuracy Comparison of Proposed Method with Existing Methods.
The contributions of the proposed method are as follows
(i) The right bits of each image pixel are segmented into two group viz., MSB and LSB
(ii) Only four MSB bits of each pixel are considered for feature extraction which leads to a
reduction in the total number of bits of an image to 50%, that increases speed of computation
and reduction in hardware complexity
(iii) The final features are extracted from an image with only four MSB bits for each pixel using
DWT
(iv) The LL band coefficients of DWT are considered as final features.
(v) The ED and ANN matching techniques are used to test the performance of the proposed
method
(vi) The proposed method is efficient for real time implementation as total number bits in an
image compressed to 50% and dimensions of an image is also reduced by considering only
LL band of DWT.
6. CONCLUSION
Face is a powerful biometric trait to recognize persons for security issues. In this paper, MSB based
face recognition using compression and dual matching technique is proposed. The eight bits of each
pixel of original face images are segmented into two part viz., MSB and Least Significant Bits (LSB).
The MSB's are having significant information of an image even without LSB's. Hence MSB's are
considered further to extract final features using LL-DWT band coefficients. The matching techniques
such as ED and ANN are used to compare the database and test image features to test the
performance of the proposed method. It is observed that, the performance of the proposed method is
better than existing methods. In future, the proposed method can be implemented using the real time
systems as the complexity of the hardware architecture is less.
7. REFERENCES
[1] Abdi. H. and Williams, "Principal Component Analysis", Wiley Interdisciplinary Reviews:
Computational Statistics, vol. 2, issue 4, pp. 433–459, 2010.
[2] T.Ahonen, A. Hadid, M. Pietikainen, "Face Description with Local Binary Patterns: Application to
Face Recognition", IEEE Transactions on Pattern Analysis and Machine Intelligence, vol. 28, no.
12, pp. 2037-2041, Dec 2006.

[3] Sujatha B M, K Suresh Babu, K B Raja and Venugopal K R “Hybrid Domain based Face
Recognition using DWT, FFT and Compressed CLBP” International Journal of Image
processing, Vol 9, Issue 5, pp 283-303, 2015.
[4] Meihua Wang, Hong Jiang and Ying Li, “Face Recognition based on DWT/DCT and SVM”, IEEE
International Conference on Computer Application and System Modeling, 2010, pp v3_ 507-
v3_510.
[5] Varsha Patil andTanuja Sarode, “Image Hashing Using DWT-CSLBP”, Journal of Computers,
vol. 14, no. 3, pp. 210-222, 2019.
[6] Ravi J and K B Raja, “Hybrid Domain Based Face Recognition System,” International Journal of
Advanced Networking and Applications, vol. 03, pp.1402-1408. 2012.
[7] Vikash, and Rajesh Parihar, “Comparison of DFT and Wavelet based Image Modification
Techniques”, International Journal of Computer Science and Mobile Computing, vol 4, issue 8,
pp 61-65, Aug 2015.
[8] Thamizharasi Ayyavoo and Jayasudha John Suseela “Illumination pre-processing method for
face recognition using 2D DWT and CLAHE,” International Journal of IET Biometrics, vol 7, issue
4, pp 380-390, 2018.
[9] Zied Bannour Lahaw, Dhekra Essaidani and Hassene Seddik, “Robust Face Recognition
Approaches Using PCA, ICA, LDA Based on DWT, and SVM algorithms” Forty First IEEE
International Conference on International Conference on Telecommunications and Signal
Processing (TSP), Jul 2018, pp 413-417.
[10] Govardhan Mattela and Sandeep K. Gupta, “Facial Expression Recognition Using Gabor-Mean
DWT Feature Extraction Technique” IEEE International Conference on Signal Processing and
Integrated Networks, 2018, pp 575-580.
[11] Jyothi Ravikumar, Ramachandra A C, K B Raja and Venugopal K R, “Convolution based Face
Recognition using DWT and HOG,” IEEE International Conference on Image Processing,
Computer Science and Information Technology, 2018, pp 327-334.
[12] P.S.Sumathi and D.Christopher Durairaj, “Enhanced Face Recognition using Principal
Component analysis (PCA) and Discrete Wavelet Transform”, International Journal of Pure and
Applied Mathematics, Vol. 119, Issue 12, pp. 13685-13693, 2018.
[13] Ridha Ilyas Bendjillali, Mohammed Beladgham, Khaled Merit and Abdelmalik Taleb-Ahmed,
“Improved Facial Expression Recognition Based on DWT Feature for Deep CNN”, International
Journal on the Science of Electronics and its Applications (Electronics), pp. 1-16, 2019.
[14] Taif Alobaidi and Wasfy B. Mikhael, “Face Recognition System based on Features Extracted
from Two Domains”, IEEE International Midwest Symposium on Circuits and Systems, pp. 977-
980, 2017.
[15] The ORL Database of Faces, AT & T (Olivetti) Research Laboratories, Cambridge, U.K.
[16] M. J. Lyons, M. Kamachi and J. Gyoba, "Japanese Female Facial Expressions (JAFFE),"
Database of digital images, 1997.
[17] Yale University, 'The Yale Face Database', 1997. [Online].
Available:http://cvc.cs.yale.edu/cvc/projects/yalefaces/yalefaces.html.
[18] Kuang-Chih Lee, David J. Kriegman, and Jeffrey Ho, “Acquiring Linear Subspaces for Face
Recognition under Variable Lighting,” IEEE Transactions on Pattern Analysis & Machine
Intelligence, May 2005, vol. 27, no 5, pp. 684-698.
[19] Ekenel H K, and Sankur B, “Multiresolution Face Recognition”, Elsevier Journal on Image and
Vision Computing, vol 23, issue 5, pp. 469–477, 2005.

[20] Sellahewa, H., Jassim, S.A, “Image-Quality-based Adaptive Face Recognition," IEEE
Transactions on Instrumentation and Measurement, April 2010, vol 59, issue 4, pp. 805–813.
[21] W. Yu, L. Gan, S. Yang, Y. Ding, P. Jiang, J. Wang and S. Li, “An Improved LBP Algorithm for
Texture and Face Classification”, Springer Verlag Journal on Signal, Image and Video
Processing, vol.8, no.1, pp.155-161, Dec 2014.
[22] Sharif M, Ayub K, Sattar D and Raza M., "Real Time Face Detection", Sindh University Research
Journal (Science Series), vol. 44(4), 597- 600, 2012.
[23] Bhaskar Anand and Prashant, “Face Recognition using SURF Features and SVM Classifier”,
International Journal of Electronics Engineering Research, vol 8, no.1, pp. 1-8, 2016.
[24] Riddhi A. Vyas, and S M Shah “Comparison of PCA and LDA Techniques for Face Recognition
Feature based Extraction with Accuracy Enhancement”, International Research Journal of
Engineering and Technology, vol 4, issue 6, pp 3332-3336, Jun 2017.
[25] Biao Yang, Jinmeng Cao, Rongrong Ni and Yuyu Zhang, “Facial Expression Recognition using
Weighted Mixture Deep Neural Network based on Double-Channel Facial Images”, IEEE
Access, Vol.6, pp.4630- 4640, 2018.
[26] Yuanyuan Ding, Qin Zhao, Baoqing Li, and Xiaobing Yuan, “Facial Expression Recognition from
Image Sequence based on LBP and Taylor Expansion”, IEEE Access Special Section on
Sequential Data Modeling and Its Emerging Applications, Vol. 5, pp.19409-19419, 2017.
[27] Yuqi Pan and Mingyan Jiang, “Kernel Low-Rank Embedding Dictionary Learning for Race
Rrecognition”, International Conference on Natural Computation, Fuzzy Systems and Knowledge
Discovery, 2017, pp. 2330-2333.
[28] Shih-Ming Huang and Jar-Ferr Yang, “Improved discriminant nearest feature space analysis for
variable lighting face recognition”, IEEE International Symposium on Circuits and Systems, 2013,
pp. 2984 – 2987.

MSB based Face Recognition Using Compression and Dual Matching Techniques

More Related Content

What's hot (19)

Similar to MSB based Face Recognition Using Compression and Dual Matching Techniques (20)

Recently uploaded (20)

MSB based Face Recognition Using Compression and Dual Matching Techniques