ISSN (Print) : 0974-6846

Indian Journal of Science and Technology, Vol 7(8), 1170–1175, August 2014 ISSN (Online) : 0974-5645

Improving Face Recognition Rate

with Image Preprocessing
Krishna Dharavath*, Fazal Ahmed Talukdar and Rabul Hussain Laskar
Speech and Image Processing Research Lab, Department of Electronics and Communication
Engineering, National Institute of Technology Silchar, Assam-788010, India;
[email protected]; [email protected], [email protected]

Abstract
Quality of image plays a vital role in increasing face recognition rate. A good quality image gives better recognition rate
than noisy images. It is more difficult to extract features from such noisy images which in-turn reduces face recognition
rate. To overcome problems occurred due to low quality image, pre-processing is done before extracting features from
the image. In this paper we will analyze the effect of pre-processing prior to feature extraction process with respect to the
face recognition rate. This also gives a qualitative description of various pre-processing techniques and feature extraction
schemes that were used for our analysis. The results were analyzed with the help of bar graphs. The combined method of
feature extraction (Spatial and Frequency) shows superior performance than individual feature extraction schemes. Also,
this combined method gives good recognition results even without pre-processing of the image.

Keywords: Comparative Study, Face Recognition, Frequency Feature, Image Preprocessing, Image Quality, Spatial Feature

1. Introduction at a distance and matche it against a database of images

to recognize the person. his process of recognizing a
In recent days, face recognition has gained a greater sig- person is however dependent on image capturing device
niicance as a unique bio-metric recognition method for such as video surveillance cameras. he chances of cor-
accurate authentication1–4, 17, 24, 27, 28. Many of the industries rect recognition depend on quality of image captured.
throughout the world are now trying to implement face Quality of image is dependent on eiciency of camera,
recognition based authentication system in their organi- distance between person and camera, lighting condition
zations to protect their assets. Many of the governments and whether the person looks at camera etc.13. In such
in the world also interested to have face recognition based case, pre-processing the captured image can give better
video surveillance system in the public places such as results in recognition.
airports, railway stations and bus stations etc. However Various image processing techniques can be used to
they remain unsuccessful due to poor recognition rate enhance the captured image and there increase in the
in recognizing a person in real time due to various fac- recognition rate. Image normalization, de-nosing, ilter-
tors5,12,22,23. Image quality is being one of the major factors ing, histogram equalization, image resizing and cropping
that greatly impact face recognition rate. and accurate face detection are certain techniques to
In an uncontrolled environment, a person may not enhance image quality and improve recognition rate13,18.
be aware of recognition system in which the recogni- Image pre-processing is done prior to extraction of fea-
tion process will be done automatically2. In such case, an tures from the image. Figure 1 shows the block diagram
image capturing device will automatically detect a person representation of general face recognition system. As

*Author for correspondence

 Krishna Dharavath, Fazal Ahmed Talukdar and Rabul Hussain Laskar

Figure 1. Block diagram of general Face Recognition System.

seen in Figure 1, an image will be captured using a cam-

era and fed as probe image to the recognition system. he
images then are pre-processed to enhance its quality. he
features are then extracted using suitable schemes. hese
features are then classiied using appropriate classifying
algorithm. In this paper, we discuss various image pro-
cessing techniques and three feature extraction schemes
that were used for our analysis. A comparative analysis
is also present with the help of bar graphs. he rest of the
paper is organized as follows. In section-2, various image
pre-processing techniques were discussed. In section-3,
various feature extraction schemes were discussed. In Figure 2. Block diagram of image pre-processing.
section-4 experimental results were discussed and sec-
In this experiment, we used image based approach
tion-5 concludes the paper.
which includes a window scanning technique with ixed
and dynamic mask size and feature based approach which
2. Image Preprocessing includes color segmentation11,14. Mask size is determined
Techniques empirically for better recognition rate. In feature based
approach, image is transferred from RGB color space
Figure 2 shows the basic pre-processing methods used to YCbCr color space provided Cb and Cr values satis-
in our experiment. he input color image is converted ies following conditions: 77≤Cb≤127 and 133≤Cr≤173.
into gray image. Using suitable cropping (face detection) Using dilation, erosion and morphological operations
schemes, the image is cropped and then resized to meet face is detected. Figure 3 illustrates the results obtained
the requirement. he image is then normalized to have using these three face detection schemes.
uniform intensity/gray level. Image is then iltered using
low pass ilter.
2.2 Image Resizing
2.1 Face Detection and Cropping he resulted image from various face detection schemes
Face cropping is also an important task to achieve high has been resized using nearest neighbor interpolation
recognition rate. Cropping can be done using various face method with speciied output size. Figure 4 shows origi-
detection techniques. Face detection13 involves detecting nal image and resized image.
a face from an image using complete image (image based
approach) or by detecting one or more features from the 2.3 Image Normalization
image (Feature based approach) such as nose, eyes, lips Illumination variation is one of the important challenges
etc. Face detection can also be done based on active shape in face recognition6–10. Image with uncontrolled light-
models such as locating head boundary19. ing conditions contains non-uniform contrast, i.e. the

Vol 7 (8) | August 2014 | www.indjst.org Indian Journal of Science and Technology 1171
 Improving Face Recognition Rate with Image Preprocessing

Figure 5. Image before and ater histogram equalization.

Figure 3. Block diagram of image pre-processing.

Detected faces using (a) dynamic mask (Upper row) and
(b) Fixed Mask (Lower row).

Figure 6. Image before and ater Filtering.

Figure 4. Resized image.
distribution of intensity/gray levels is not alike. To make
these levels equal or almost equal we use histogram equal-
ization technique26. Given an M x N image, Cumulative
Distribution Function (CDF) at each Pixel Value (V) is
used to ind pixels equalization value (h) over L gray lev-
els. Mathematically it is given as 28,
 CDF (V ) − CDFmin 
h=  ( L − 1)
 MXN − CDF min 
Figure 5 shows the efect of histogram equalization on
input image.

2.4 Image De-noising and Filtering Figure 7. Block diagram of feature extraction process.

Images are oten by default have Gaussian noise due to

illumination variations. To de-noise it we work on pixel 3.1 Eigen Face based Approach
based iltering techniques18. In our experiment, we used Principal Component Analysis (PCA) is used to extract
Low Pass Filter (LPF) to eliminate high frequency infor- Eigen faces12,16,17,22–24. Initially Eigen vectors are computed
mation and retain only with low frequency information. using covariance matrix derived from set of training
Figure 6 shows the diference between noisy image and images. Probe image is then projected in to the face space
iltered image. and the distance between mean Eigen face and probe
image is computed using spatial diferential operators
3. Feature Extraction Techniques such as Euclidian distance and Cosine distance.

Input image ater preprocessing fed to a feature extraction

3.2 Discrete Cosine Transform
scheme to extract features from it as shown in Figure 7.
Eigen face approach and Discrete Cosine Transform Discrete Cosine Transform (DCT) transforms spa-
(DCT) approach were used to extract features. However tial domain images into decoupled frequency domain
we have also tried combined approach using these two images7,15,24. hat means image information is converted
methods. Recognition rate of these methods were com- into DCT coeicients. hese are located in the upper let
puted with and without image pre-processing prior to corner of the DCT maximized and are the most impor-
feature extraction. tant to represent image back.

1172 Vol 7 (8) | August 2014 | www.indjst.org Indian Journal of Science and Technology
 Krishna Dharavath, Fazal Ahmed Talukdar and Rabul Hussain Laskar

3.3 Combined Approach he impact of image cropping with diferent methods

on face recognition rate is shown in Figure 9. From Figure
In this approach we used both spatial and frequency
9 it is evident that dynamic cropping and colour segmen-
domain methods Spatial domain features are extracted
tation methods shows high impact on face recognition
using spatial diferential operators and frequency
rate. Similarly, Figure 10 illustrates the impact of image
domain features are extracted using DCT. Each method
normalization on face recognition rate.
is followed by dimensionality reduction process. PCA
is used to reduce dimensionality of SDO features and
Zonal coding is used in DCT feature dimensionality 5. Conclusion
reduction28.
We have demonstrated how face recognition rate
improves with diferent image preprocessing techniques.
4. Results and Discussions Our investigation proves that there is a considerable
impact on face recognition rate especially when images
Ater extracting features from images, Linear Discriminate
Function20,21,25 is used to classify features and then com-
pared these features against available features in database.
We have investigated the impact of preprocessing tech-
niques on recognition rate on our Speech and Image
Processing Research Lab (SIPRL) Database29. It includes
a total of 35 subjects. Each subject has various pose, illu-
mination, occlusion and expressions which lead to an
average of 19 images per subject. here are some post
graduate students, research scholars and teaching fac-
ulties were involved in this project. We have used semi
controlled environment for subject’s registration. We have
varied lighting conditions also.
We have crated three diferent testing sets which vary Figure 8. Average improvement in FR rate.
in resolution and illumination. Each testing set consist
three subject’s images. First we have used Method-1 to
analyze the impact of pre-processing on recognition
rate. Table 1 illustrates the efect of pre-processing on
face recognition rates using three approaches. It shows
that method-1 and method-2 has greater impact on
recognition rate than method-3. here is an average of
66.32% improvement in face recognition rate in case of
method-1 and is 62.12% in case of method-2. However
in method-3, it has a little lesser impact on face recogni-
tion rate, only 48% as shown in Figure 8. It is because
of combination of both spatial and frequency domain
Figure 9. Impact of image cropping on FR rate..
features.
Table 1. Impact of image preprocessing on face recognition rate
Face Recognition rate
Method-1 Method-2 Method-3
Test Set Raw Preprocessed Raw Preprocessed Raw Preprocessed
TS1 22.62% 68.77% 18.92% 69.85% 27.23% 79.08%
TS2 29.38% 85.54% 41.85% 89.09% 62.77% 92.62%
TS3 19.50% 58.00% 25.24% 68.15% 40.73% 79.69%

Vol 7 (8) | August 2014 | www.indjst.org Indian Journal of Science and Technology 1173
 Improving Face Recognition Rate with Image Preprocessing

reduction approach for face recognition. Signal Processing.

2008; 88:2604–09.
8. Dang HL, Kin ML, Lan SS. Illumination Invariant Face
Recognition. Pattern Recognition. 2005; 38:1705–16.
9. Ognjen A, Roberto C. A methodology for rapid illumi-
nation-invariant face recognition using Image processing
ilters. Computer Vision and Image Understanding. 2009;
113:159–71.
10. Xudong X, Kin ML . an eicient illumination normalization
method for face recognition. Pattern Recognition Letters.
2006; 27:609–17.
11. Zhiming L, Chengjun L. A hybrid color and frequency fea-
Figure 10. Impact of normalization on FR rate. tures method for face recognition. IEEE Image Processing.
2008; 17:1975–80.
12. Weilong C, Meng JE, Shiqian W. PCA and LDA in DCT
are of diferent size and have lighting variations. Noisy Domain. Pattern Recogn Lett. 2005; 26(15): 2474–82.
images show a greater impact on recognition rate. In such 13. Paul V, Michael JJ. Robust Real-Time Face Detection.
cases image has to be preprocessed appropriately. We International Journal of Computer Vision. 2004; 57:
have also tested a combined approach for feature extrac- 137–54.
tion which has shown superior results. his approach has 14. Chai D, Ngan KN. Face segmentation using skin-color map
also shown good results even without preprocessing prior in videophone applications. IEEE Trans Circ Syst Video
to the feature extraction. It can be concluded that, prepro- Tech. 1999; 9:551–64.
cessing should be done prior to feature extraction from 15. Hajiarbabi M, Askari J, Sadri S, Saraee M. Face rec-
face images. Future scope of this study is to analyze the ognition using discrete cosine transform plus linear
same system with diferent feature extraction schemes as discriminant analysis. Proceedings of the World Congress
well as classiiers. on Engineering. 2007; 2165:652–55.
16. Wang L, Yan, Z, Jufu F. On the Euclidean Distance of
Images. IEEE Trans Pattern Anal Mach Intell. 2005; 27:
6. References 1334–39.
1. Anil KJ, Arun R, Salil P. An Introduction to Biometric 17. Matthew AT, Alex PP. Face recognition using Eigen faces.
Recognition. IEEE Trans Circ Syst Video Tech. 2004; Proceedings of IEEE Computer Society Conference
14(1):4–20. on Computer Vision and Pattern Recognition. 1991;
2. Krishna D, Fazal AT, Rabul HL. Study on Biometric authen- 586–91.
tication systems, challenges and future trends. Proc of IEEE 18. Peng DL, Xue A. Degraded image enhancement with appli-
International Conference on Computational Intelligence cations in robot vision. Proceedings of IEEE International
and Computing; 2013. p. 1–7. Conference on Systems, Man and cybernetics. 2005;
3. Andrew W, Ruud MB. Face Recognition and its applica- 2:1837–42.
tions. Chapter 4. Springer; 2002. 19. Omid,K, Hamid RS, Ahmad JM. An overview on model
4. Tanaya M, Jonathan WQM. Face Recognition using based approaches in face recognition, 7th WSEAS
Curvelet based PCA. Signal Processing. 2009; 89:2345–53. International Conference on Artiicial Intelligence,
5. Fernandes SL, Bala J. Performance analysis of linear appear- Knowledge Engineering and Data Bases; 2008.
ance based algorithms for face recognition. Int J Comput p. 109–15.
Trends Tech. 3, 258-267; 2012. 20. Haralick, Robert M. Textural features for image classi-
6. Taiping Z, Yuan YT, Bin F, Shang Z, Xiaoyu L. Face ication. IEEE Trans Syst Man Cybern C Appl Rev. 1973;
Recognition under varying illumination using gradi- 3:610–21.
entfaces. IEEE Transactions on Image Processing. 2009; 21. Lihong Z, Xiangjian H. Classiication techniques in pattern
18:2599–606. recognition. International Conference in Central Europe
7. Dattatray VJ, Raghunath SH. Radon and discrete cosine on Computer Graphics, Visualization and Computer
transforms based feature extraction and dimensionality Vision; 2005. p. 77–8.

1174 Vol 7 (8) | August 2014 | www.indjst.org Indian Journal of Science and Technology
 Krishna Dharavath, Fazal Ahmed Talukdar and Rabul Hussain Laskar

22. Delac K, Grgic M, Grgic S. Independent Comparative Study Approach for Face Recognition. IEEE Trans Image Process.
of PCA, ICA, and LDA on the FERET Data Set. Int J Imag 2007; 16(1):57–68.
Syst Tech. 2002; 15(5):252–60. 27. Scott WL. Block-level discrete cosine transforms coei-
23. Martinez AM, Kak AC. PCA versus LDA. IEEE Trans cients for autonomic face recognition [Ph.D. Dissertation].
Pattern Anal Mach Intell. 2001; 23(2):228–33. USA: Louisiana State University; 2003.
24. Stan ZL, Anil KJ. Handbook of face recognition,. Chapter 1, 28. Kamal AA. Face recognition in uncontrolled indoor envi-
Springer; 2004. ronment [MS Dissertation], AUS; 2013 Jun.
25. Yang J, Yu Y, Kunz W. An Eicient LDA algorithm for 29. Krishna, D. Speech and Image Processing Research Lab
face recognition. International Conference on Control, (SIPRL) Database. Paper presented at: he research schol-
Automation, Robotics and Vision; 2000. p. 1–6. ar’s meeting, Electronics and Communication Engineering
26. Zhang B, Shan S, Chen X, Gao W. Histogram of Gabor department of National Institute of Technology Silchar;
Phase Patterns (HGPP). A Novel Object Representation 2013 July 16; Silchar, India.

Vol 7 (8) | August 2014 | www.indjst.org Indian Journal of Science and Technology 1175