PERFORMANCE EVALUATION OF BLOCK-SIZED ALGORITHMS FOR MAJORITY VOTE IN FACIAL RECOGNITION

International Journal of Artificial Intelligence and Applications (IJAIA), Vol.14, No.5, September 2023
DOI:10.5121/ijaia.2023.14501 1
PERFORMANCE EVALUATION OF BLOCK-SIZED
ALGORITHMS FOR MAJORITY VOTE IN
FACIAL RECOGNITION
Andrea Ruiz-Hernandez , Jennifer Lee, Nawal Rehman, Jayanthi Raghavan
and Majid Ahmadi
Department of Electrical and Computer Engineering, University of Windsor,
Windsor, Canada.
ABSTRACT
Facial recognition (FR) is a pattern recognition problem, in which images can be considered as a matrix of
pixels.There are manychallenges that affect the performance of face recognitionincluding illumination
variation, occlusion, and blurring. In this paper,a few preprocessing techniques are suggested to handle the
illumination variationsproblem. Also, other phases of face recognition problems like feature extraction and
classification are discussed. Preprocessing techniques like Histogram Equalization (HE), Gamma Intensity
Correction (GIC), and Regional Histogram Equalization (RHE) are tested inthe AT&T database. For
feature extraction, methods such as Principal Component Analysis (PCA), Linear Discriminant Analysis
(LDA), Independent Component Analysis (ICA), and Local Binary Pattern (LBP) are applied. Support
Vector Machine (SVM) is used as the classifier. Both holistic and block-based methods are tested using the
AT&T database. For twelve different combinations of preprocessing, feature extraction, and classification
methods, experiments involving various block sizes are conducted to assess the computation performance
and recognition accuracy for the AT&T dataset.Using the block-based method, 100% accuracy is achieved
with the combination of GIC preprocessing, LDA feature extraction,and SVM classification using 2x2
block-sizingwhile the holistic method yields the maximum accuracy of 93.5%. The block-sized algorithm
performs better than the holistic approach under poor lighting conditions.SVM Radial Basis Function
performs extremely well on theAT&Tdataset for both holistic and block-based approaches.
KEYWORDS
Principal Component Analysis (PCA), Linear Discriminant Analysis (LDA), and Independent Component
Analysis (ICA).
1. INTRODUCTION
Biometric recognition includes various techniques like fingertips, face recognition, and iris
methods. Except FR, other methods demand an individual’s involvement to access the system.
Among such methods, face recognition is a powerful technique that can be carried out in a covert
manner [6]. FR has undergone significant advancements in recent years and has been widely
implemented in various applications including access control systems, driving licenses [2],
passport authentication, smartphones, public safety [3] criminal identification [1], and network
security [4].However, the performance of face recognition is affected by many factors like
occlusion, aging, similar faces, expression variation, and resolution [5]. These poor conditions
significantly impact the facial recognition system resulting in inaccurate identification. Face
imagesare often affected by illumination variation. Even with the best face recognition systems,
the recognition accuracy may beaffected by illumination variation [17].

2
Image preprocessing is a technique to convert raw image information into perfect data, as the raw
image may have noisy, incomplete, and/or unpredictable values [15]. There are many
preprocessing algorithms available in the literature to improve the captured image in turn
boosting the recognition rate. Some popular techniques include but are not limited to Image
normalization, de-nosing, HistogramEqualization, image resizing, and cropping. Image pre-
processing is performed before the feature extraction phase [16].
Feature extraction is one of the important phases of face recognition. Numerous feature
extraction methods are applied to extract features of facial images [18]. Some of the extensively
applied feature analysis methods are Principal Component Analysis (PCA), Linear Discriminant
Analysis (LDA), Local Binary Pattern (LBP), and Independent Component Analysis (ICA).
The output of the feature extractor is fed as the input to the classifier. Facial images are classified
by applying a wide range of classifiers to the respective classes of faces [19]. A few examples of
popular classifiers are Random Forest (RF), K-Nearest Neighbor (KNN), and Support Vector
Machine (SVM)
2. RELATED WORK
In recent times, automatic face recognition systems obtained the highest accuracy incontrolled
environments. Yet, there are challenges that need to be addressed in uncontrolled environments
[20]. Illumination variation is one of the greatchallenges that must be handled properly to
achievehigher accuracy in the FR system. Various approaches have been recommended to resolve
this problem. In this paper, a few preprocessing methods to handle the illumination variation
problem are suggested [20].
Histogram Equalization (HE) is one of the widely applied normalization methods in
preprocessing. Equalizing a histogram is to extend and reorganize the original histogram using
the whole range of discrete levels of the image [10]. The histogram equalization does not work
well when there is a non-uniform illumination variation.
For the image I (x, y) with discrete k. gray values, histogram is defined by the probability of
occurrence of the gray level I [27], given by Equation (1) as follows:
P (i) = ni / N (1)
Where i∈0, 1…k – 1 grey level and N is total number of pixels in the image.
HE is applied to the entire image and RHE is performed on smaller regions within the image. The
image is divided into regions and histogram equalization is applied separately to each region.
This method enhances each region’s contrast independently [8], which preserves the details and
prevents noise that can occur with global HE
Another type of HE recommended in literature is namely Contrast Limited Adaptive Histogram
Equalization (CLAHE), which works extremely well under local lighting varieties [9].
GIC is a preprocessing technique that can be applied to control the overall brightness of the
image. It is a nonlinear approach shown in Fig.1.While usual image normalization techniques like
scalar multiplication and addition/subtraction perform linear operations on individual pixels [13],
GIC [12] shown in Fig. 1 is applied to correct the differences between the content captured by the
camera the manner in which the display device displays, content and the way the human visual

3
system handleslight. GIC transforms gray-level I to gray-level 𝐼1/𝛾
andt is given by the following
equation (2). But Gamma Correction does not eliminate problems like shading effects [10].
I=𝐼1/𝛾
(2)
Figure 1.Gamma Intensity Correction
The adaptive Gamma correction method was recommended [14] to overcome the limitations of
simple gamma correction. It partitions the value of the gray-scale image into three
partshighlights, transition, and shadow, and then creates the nonlinear association between the
gray-scale value and the Gamma value
The feature extraction process is the next step in face recognition. By applying the LBP operator,
the features of the image are extracted by partitioning into small areas called blocks and then the
binary pattern histograms are extracted [11]. T. Ojala et al. [7] introduced the original LBP, which
is a powerful extractor to represent image texture.
Not all the features extracted from feature extraction are useful data. To process the huge
extracted dataeffectively, its dimensionality must be decreased. Basically, dimensionality
reduction converts data with higher dimensionality to a lower dimensionality through a linear
transformation.Principle component Analysis (PCA) is the best example ofa dimensionality
reduction technique. The reduced meaningful data representation must have a dimensionality that
corresponds to the intrinsic dimensionality of the data [29].The intrinsic dimensionality of data
represents the number of variables needed in a minimal representation of the data [30].
During the conversion, most representative data is kept and the noisy, redundant data is removed.
Because of this feature, PCA is extensively applied as a dimensionality reduction tool in face
recognition systems [24].In this paper, the optimal number of features required for each feature
extraction method to get the highest accuracy is discussed.
Linear Discriminant Analysis (LDA) explicitly aims to model the difference between the data of
various classes. LDA performs extremely wellin face recognition applications, by decreasing the
intra-class variation that occurs between individuals in the same class due to illumination
variation and increasing the inter-classvariation in appearance due to differences in identity [25].
Independent Component Analysis (ICA) is a general form of PCA [25]. PCA optimizes second-
order data, which is the covariance matrix. But ICA optimizes higher-order data like kurtosis. For
face recognition applications, it is necessary to understand the higher-order relationship between
the pixels.ICA performs better than PCA in FR [26].
SVMs fall under the category of maximum margin classifiers. SVM carries out classification
between two classes by detecting a decision margin that has thegreatest distance to the nearest

4
points in the training set which are known as support vectors [31]. SVMs are effective binary
classifiers that search the feature space for a decision hyperplane that maximizes the difference
between two classes (-1 and 1) [21][22][23]. Although SVM handles non-linearly separable data
using kernelfunctions, SWMs are most successful when the data is linearly separable. However,
SVMs can be computationally intensive when working with large datasets and non-linear
functions.
Hongshuai Zhang etl [32] recommended a face recognition model that makes use of the LBP
feature and CNN. LBP is applied to transform the image to a feature map, then the LBP feature
map is applied to CNN as the input train CNN. The recommended combination of LBP and CNN
performed better in terms of accuracy, sensitivity, and specificity compared to applying CNN
alone.
Dadi, Harihara Santosh, and others suggested a face recognition method, in which the Histogram
of Oriented Gradient is applied as a feature extractor and SVM as s classifier[33].Extracted
features from HOG are fed to the Support Vector Machine classifier. The system achieved a good
accuracy of 92% in the Yale database
3. EXPERIMENT
Experiments are conducted in the AT&T database, which consist of 400 images [28]shown in
Fig.2 using the holistic and block-based methods.In this experiment, 80% of the data was used
for training and the remaining 20% for testing. Experiments are conducted to find the optimal
number of features that would result in the highest accuracy rate and the best parameter values
for various preprocessing techniques.
Figure 2. Sample Images from AT&T database.
3.1. Holistic Method
The holisticmethod shown in Fig.3 accepts the entire facial image as the input. In this method, the
preprocessing methods are applied to the complete image. Then the image is block sized.
Figure 3.Implementation of the holistic approach

5
HE, GIC, and RHE are commonly used methods to enhance image quality before features are
extracted. In HE, the overall illumination and visibility of details in an image are enhanced. This
is achieved by redistributing the intensity value across the entire range. In the given dataset, the
intensity values range from 0 to 255, which represents the pixel’s illumination. The first stage in
HE is to plot the histogram of the image, which is the frequency of each intensity value. Then, the
CumulativeDistribution unction (CDF) of the histogram was calculated. The CDF is the
probability of each intensity value occurring in the given image. After, the CDF was normalized
to cover the intensity values ranging from 0 to 255, anda mapping function was created. Then,
new intensity values were assigned to the original input image based on the mapping function.
GIC was applied to the input images. The overall brightness of the image can be controlled by the
gamma value (γ). When γ is less than 1, the image appears brighter because the low-intensity
values are heightened. In contrast, when γ is greater than 1, the image appears darker as higher
intensity values are amplified. The three gamma values, 0.7, 1.5, and 2.2, were tested and the
results are given in Table 5.
For feature extractionPCA, LDA, ICA, and LBP methods are applied. Using PCA, the data points
with the greatest variance are extracted, which also contain the most useful information. This
reduces the computational burden significantly and eliminates any redundancy of features.When
the number of feature vectors increases, the computational burden also increases but it givesmore
accuracy.
The other hyperparameters, namely the kernel (Radial Bias Function (RBF) in this case) and
gamma were found through the trial-and-error method. To decide the best kernel, all the basic
kernel functions like Linear, Polynomial, and Radial Basis Functions (RBF) are tested. Based on
the results, RBF is chosen as the best kernel function for all cases. To find the best gamma value,
an arbitrary value was set first, and from multiple trial-and-error iterations, the best gamma value
which gave the best performance for each case was found. The parameters used for the Holistic
method for all the preprocessing methods like HE, GIC, and RHE are given below in Table 1.
The accuracy rates are given in Table 2
SVM is applied as a classifier in this experiment. SVMs are binary classifiers meant to optimally
separate two sets of data by using straight lines. There are two general approaches tofinding the
optimal hyperplane, the first being a maximum-margin hyperplane, in which the aim is to place
the hyperplane “in the middle” of both data clusters to maximize the distance from the
hyperplane to both clusters [15]. A soft margin differs in that it allows for some misclassifications
to account for anomalous data points [15]. The number of misclassifications allowed is generally
tuned using the hyperparameter C. This type of margin was selected for the SVM used in this
experimentas indicated in the last column of Table 1.

6
Table 1: Parameters for classification using the whole image
Preprocessing Feature
Extraction
Kernel Gamma C
HE LDA RBF 35 1
HE PCA RBF 15 1
HE LBP RBF 0.007 1
HE ICA RBF 15 1
GIC LDA RBF 35 1
GIC PCA RBF 15 1
GIC LBP RBF 0.05 1
GIC ICA RBF 15 1
RHE (4x4) LDA RBF 15 1
RHE (4x4) PCA RBF 20 1
RHE (4x4) LBP RBF 0.007 1
RHE (4x4) ICA RBF 10 1
RHE (4x4) PCA 85.5
RHE (4x4) LBP 76.5
RHE (4x4) ICA 90.75
Table 2:Accuracy results for facial recognition using whole image for classification (holistic approach)
Preprocessing Feature Extraction Accuracy (%)
HE LDA 86.5
HE PCA 90
HE LBP 73
HE ICA 90
GIC LDA 91.75
GIC PCA 93.5
GIC LBP 79.25
GIC ICA 93.25
RHE (4x4) LDA 71.75
3.2. Block-Based Method
It is difficult to achieve higher accuracy using a holistic approach, as it is affected by illumination
variation. The block-based method shown in Fig.4 is different from the holistic approach, as the
input image was divided into smaller sub-images/blocks and each sub-image was processed
separately. The separate blocks would be recombined at the end with their classifier scores
summed up to determine the class of the input image. The general steps for the block-sizing
approach are given as follows.
A specialized preprocessing technique called RHE is applied in the block-based method. It is an
extension of the traditional histogram equalization that applies to local regions of the image.
Figure 4.Implementation of Block-sized Method

7
Instead of implementing HE globally to the entire image, RHE is performed on smaller regions
within the image. The image is divided into N-by-N non-overlapping regions and histogram
equalization is applied separately to each region. This method enhances each region’s contrast
independently, which preserves the details and prevents over-amplification of noise that can
occur with global HE. The number of blocks (N-by-N) is a variable parameter. The experiment is
conducted for 2x2, 4x4, and 8x8 block sizes,and the accuracies are tested for each block size
given in Table 6.
For the GIC preprocessing technique, the most optimal gamma value was assigned to each block,
depending on the mean pixel intensity.Three gamma correction factors with the values 1.8 (low),
1.5 (medium), and 2.2 (high) are chosen based on the measured brightness of the region by trial-
and-error method.
In the block-sized approach, feature extraction was applied to each block as an individual image.
The optimal number of features necessary to achieve the highest accuracy is the same for both
holistic and block-sized methods. Results are given in Table 3. Support Vector Machine (SVM) is
applied as a classifier for block-sized methods also. Parameters for classifying different values of
Regional Histogram Equalization (RHE) for different block sizes is given in Table 4. Optimal
Parameters for preprocessing for both Holistic and Block-sized Approaches are given in Table 5.
Accuracy ratesof the block-sized approach for various block sizes are given in Table 6.
Parameters for classification using the block-sized approach are given in Table 7.
Table 3: Parameters for feature extraction for both Holistic and Block-sized Approaches
Method Number of
Components
Linear Discriminant Analysis (LDA) 4
Principal Component Analysis (PCA) 9
Local Binary Pattern (LBP) 8 (neighbor pixels)
Independent Component Analysis (ICA) 12
The performances of different block sizes for RHE for each feature extraction method are
evaluated. The final parameters used in those tests are shown in Table 4
Table 4: Parameters for classifying different values of RHE block sizes
RHE Block Size Feature Extraction Gamma
2x2 LDA 20
2x2 PCA 10
2x2 LBP 0.007
2x2 ICA 10
4x4 LDA 15
4x4 PCA 20
4x4 LBP 0.007
4x4 ICA 10
8x8 LDA 12
8x8 PCA 10
8x8 LBP 0.007
8x8 ICA 10

8
Table 5: Parameters for preprocessing for both Holistic and Block-sized Approaches
Method Parameter
Regional Histogram Equalization (RHE) Block Size = 4
Gamma Intensity Correction (GIC) Gamma = 2.2 (holistic)
Gamma = 1.2, 1.5, 1.8 (block-sized)
Table 6: Accuracy results for facial recognition using the block sizing approach for classification using 2x2,
4x4, and 8x8 block sizes
Preprocessing Feature Extraction Accuracy (%)
2x2 4x4 8x8
HE LDA 98.25 100 100
HE PCA 95.75 97.25 96
HE ICA 93.75 94 93.25
GIC LDA 100 100 100
GIC PCA 95 97.25 95.25
GIC ICA 94 96.5 994.5
Table 7: Parameters for classification using the block-sized approach
Preprocessing Feature Extraction Gamma
2x2 4x4 8x8
HE LDA 10 10 35
HE PCA 10 10 15
HE ICA 10 10 15
GIC LDA 15 15 20
GIC PCA 15 25 20
GIC ICA 10 30 15
The performances of different block sizes for RHE using the holistic approach were also
evaluated and given in Table 8
Table 8: Accuracy results using different block sizes of RHE for the whole image classification
RHE Block Size Feature Extraction Accuracy (%)
2x2 LDA 85
2x2 PCA 85
2x2 LBP 75.75
2x2 ICA 88.75
4x4 LDA 71.75
4x4 PCA 85
4x4 LBP 76.5
4x4 ICA 90.75
8x8 LDA 63.75
8x8 PCA 75
8x8 LBP 78.25
8x8 ICA 78.75

9
4. CONCLUSIONS
In this paper, both the holistic approach and block-based approach are examined to accurately and
efficiently recognize individuals under poor lighting conditions. The experiments are
conductedfor twelve different combinations of preprocessing, feature extraction, and
classification methods. Extensive experimentswere carried out for block-basedmethods with
different block sizes to assess the computation performance and recognition accuracy for the
AT&T dataset. Another aspect of the research was combining multiple preprocessing techniques
on both the entire image, as well as the blocks of an image to evaluate how the performance is
affected. The goal wasto find the facial recognition model that performed better than the others
and remained less of a computational burden.
The combination with a success rate of 100% was achieved using GIC preprocessing along with
LDA feature extraction and SVM classification, with 2x2 block-sizing. For this combination, all
individuals were successfully recognized by the trained model. Furthermore, the 2x2 block-sizing
has a shorter runtime than any other block size, combined with the faster computation of GIC,
LDA, and SVM
PCA may not be the most optimal choice as a feature extraction method for a dataset that contains
illumination variation. LDA is often used and preferred because of its simplicity and accuracy.
Experimental resultsprovethat Radial Basis Function (RBF) kernel SVMs perform best on the
AT&T facedataset.
The accuracy rate of the holistic approach is between 80% to 90%. Theaccuracy rate of the block-
sizing approach, which uses the same combinations, is much higher than the holistic approach
around 100%. The block-sizing approach performs much better than the holistic approach under
poor illumination conditions.In future work, we plan to address other challenges in face
recognition like low resolution.
REFERENCES
[1] Kortli, Y., Jridi, M., Al Falou, A., & Atri, M. (2020). Face recognition systems: A
survey. Sensors, 20(2), 342.
[2] Singh, Shilpi, and S. V. A. V. Prasad. "Techniques and challenges of face recognition: A critical
review." Procedia computer science 143 (2018): 536-543.
[3] Hu, Yongmei, et al. "The development status and prospects on the face recognition." 2010 4th
International Conference on Bioinformatics and Biomedical Engineering. IEEE, 2010.
[4] Li, Lixiang, Xiaohui Mu, Siying Li, and Haipeng Peng. "A review of face recognition
technology." IEEE Access 8 (2020): 139110-139120.
[5] Rusia, Mayank Kumar, and Dushyant Kumar Singh. "A comprehensive survey on techniques to
handle face identity threats: challenges and opportunities." Multimedia Tools and Applications 82,
no. 2 (2023): 1669-1748.
[6] Taskiran, Murat, Nihan Kahraman, and Cigdem Eroglu Erdem. "Face recognition: Past, present and
future (a review)." Digital Signal Processing 106 (2020): 102809.
[7] T. Ojala, Pietikinen and M enp , (2002) "Multi resolution gray-scale and rotation invariant texture
classification with local binary patterns", IEEE Transaction on Pattern Analysis and Machine
Intelligence, pp 971-987.
[8] Musa, Purnawarman, Farid Al Rafi, and MissaLamsani. "A Review: Contrast-Limited Adaptive
Histogram Equalization (CLAHE) methods to help the application of face recognition." 2018 third
international conference on informatics and computing (ICIC). IEEE, 2018.

10
[9] R. P. Dahake1, M. U. Kharat and S.V. Gumaste Comparative Study of Illumination Pre-processing
Techniques using Histogram Equalization and its Application in Face Recognition,
Biosc.Biotech.Res. Comm. Special Issue Vol 13 No 14 (2020) Pp-394-403
[10] S. Anila and N. Devarajan, “Preprocessing Technique for Face Recognition Applications under
Varying Illumination Conditions Preprocessing Technique for Face Recognition Applications under
Varying Illumination Conditions,” Global Journal of Computer Science and Technology Graphics &
Vision, vol. 12, 2012, Available: https://globaljournals.org/GJCST_Volume12/2-Preprocessing-
Technique-for-Face-Recognition.pdf
[11] Rouhi, R., Amiri, M., &Irannejad, B. (2012). A review on feature extraction techniques in face
recognition. Signal & Image Processing, 3(6), 1.
[12] Chapter 4 - Digital picture formats and representations David R. Bull, Fan Zhang,in Telligent Image
and video Compression (Second edition) 2021
[13] Shabana, D. F., Badithala, S., Daggupati, S., Chevala, R., & Raj, K. (2020). an Image Enhancement
Algorithm Using Gamma Correction By Swarm Optimization. Int Res J Eng Technol, 7(9).
[14] Zhang, Xue-Wen, et al. "A Method of Continuous Nonlinear Gamma Correction." 2nd Annual
International Conference on Electronics, Electrical Engineering and Information Science (EEEIS
2016). Atlantis Press, 2016.
[15] Chaki, Jyotismita, and Nilanjan Dey. A beginner’s guide to image preprocessing techniques. CRC
Press, 2018.
[16] Dharavath, Krishna, Fazal Ahmed Talukdar, and Rabul Hussain Laskar. "Improving face recognition
rate with image preprocessing." Indian Journal of Science and Technology 7.8 (2014): 1170-1175.
[17] Wei, Pengcheng, et al. "Research on face feature extraction based on K-mean algorithm." EURASIP
Journal on Image and Video Processing 2018.1 (2018): 1-9.
[18] WW Bledsoe, The model method in facial recognition, vol 15 (Panoramic Research Inc, Palo Alto,
1966), p. 47
[19] Dino, H. I., & Abdulrazzaq, M. B. (2019, April). Facial expression classification based on SVM,
KNN and MLP classifiers. In 2019 International Conference on Advanced Science and Engineering
(ICOASE) (pp. 70-75). IEEE.
[20] L. Zhichao and E. M. Joo, "Face Recognition under Varying Illumination," New Trends in
Technologies: Control, Management, Computational Intelligence and Network Systems, 2010.
[21] D. Xi, I. T. Podolak, and S.-W. Lee, “Facial component extraction and face recognition with support
vector machines,” Proceedings of Fifth IEEE International Conference on Automatic Face Gesture
Recognition, 2002.
[22] Guodong Guo, S. Z. Li, and Kapluk Chan, “Face recognition by support Vector Machines,”
Proceedings Fourth IEEE International Conference on Automatic Face and Gesture Recognition
(Cat. No. PR00580).
[23] M. O. Faruqe and M. A. Hasan, “Face recognition using PCA and SVM,” 2009 3rd International
Conference on Anti-counterfeiting, Security, and Identification in Communication, 2009.
[24] Peng, P., Portugal, I., Alencar, P. and Cowan, D., 2021. A face recognition software framework
based on principal component analysis. Plos one, 16(7), p.e0254965.
[25] Bhattacharyya, S. K., & Rahul, K. (2013). Face recognition by linear discriminant
analysis. International Journal of Communication Network Security, 2(2), 31-35.
[26] Stone, J. V. (2002). Independent component analysis: an introduction. Trends in cognitive
sciences, 6(2), 59-64.
[27] Raghavan, Jayanthi, and Majid Ahmadi. "Performance Evaluation of Weighted Entropy Based
Fusion Technique for Face Recognition with Different Pre-processing Techniques." Advances in
Information and Communication: Proceedings of the 2021 Future of Information and
Communication Conference (FICC), Volume 2. Springer International Publishing, 2021.
[28] “The Database of 31a GTDLBench,
https://gitdisl.github.io/GTDLBench/datasets/att_face_dataset/#:~:text=The%20AT%26T%20face%
20dataset%2C%20%E2%80%9C(,April%201994%20at%20the%20lab. (accessed Jul. 31, 2023).
[29] Van Der Maaten, Laurens, Eric O. Postma, and H. Jaap van den Herik. "Dimensionality reduction: A
comparative review." Journal of Machine Learning Research 10.66-71 (2009): 13.
[30] K. Fukunaga. Introduction to Statistical Pattern Recognition. Academic Press Professional, Inc., San
Diego, CA, USA, 1990.

11
[31] Heisele, Bernd, Purdy Ho, and Tomaso Poggio. "Face recognition with support vector machines:
Global versus component-based approach." In Proceedings Eighth IEEE International Conference
on Computer Vision. ICCV 2001, vol. 2, pp. 688-694. IEEE, 2001.
[32] Zhang, Hongshuai, Zhiyi Qu, Liping Yuan, and Gang Li. "A face recognition method based on LBP
feature for CNN." In 2017 IEEE 2nd Advanced Information Technology, Electronic and Automation
Control Conference (IAEAC), pp. 544-547. IEEE, 2017.
[33] Dadi, Harihara Santosh, and GK Mohan Pillutla. "Improved face recognition rate using HOG
features and SVM classifier." IOSR Journal of Electronics and Communication Engineering 11, no.
4 (2016): 34-44.

PERFORMANCE EVALUATION OF BLOCK-SIZED ALGORITHMS FOR MAJORITY VOTE IN FACIAL RECOGNITION

More Related Content

Similar to PERFORMANCE EVALUATION OF BLOCK-SIZED ALGORITHMS FOR MAJORITY VOTE IN FACIAL RECOGNITION (20)

Recently uploaded (20)

PERFORMANCE EVALUATION OF BLOCK-SIZED ALGORITHMS FOR MAJORITY VOTE IN FACIAL RECOGNITION