Effective face feature for human identification
1 like•285 views
The document discusses a new method for face feature extraction aimed at improving human identification accuracy using digital images. It highlights the challenges of face recognition, particularly variations in poses and expressions, and presents a technique that combines geometric and texture features obtained through Haar wavelet transform. Experimental results demonstrate the effectiveness of the proposed method in reducing intra-person variation while enhancing inter-person differentiation for robust face recognition.
![IJRET: International Journal of Research in Engineering and Technology eISSN: 2319-1163 | pISSN: 2321-7308
__________________________________________________________________________________________
Volume: 03 Issue: 04 | Apr-2014, Available @ http://www.ijret.org 117
EFFECTIVE FACE FEATURE FOR HUMAN IDENTIFICATION
S. Adebayo Daramola1
, Tiwalade Odu2
, Olujimi Ajayi3
1
Senior Lecturer, Department of Electrical and Information Engineering, Covenant University Ota, Ogun State, Nigeria
2
Assistant Lecturer, Department of Electrical & Information Engineering, Covenant University Ota, Ogun State, Nigeria
3
Post Graduate Student, Department of Electrical & Information Engineering, Covenant University Ota, Ogun State,
Nigeria
Abstract
Face image is one of the most important parts of human body. It is easily use for identification process. People naturally identify one
another through face images. Due to increase rate of insecurity in our society, accurate machine based face recognition systems are
needed to detect impersonators. Face recognition systems comprise of face detector module, preprocessing unit, feature extraction
subsystem and classification stage. Robust feature extraction algorithm plays major role in determining the accuracy of intelligent
systems that involves image processing analysis. In this paper, pose invariant feature is extracted from human faces. The proposed
feature extraction method involves decomposition of captured face image into four sub-bands using Haar wavelet transform thereafter
shape and texture features are extracted from approximation and detailed bands respectively. The pose invariant feature vector is
computed by fusing the extracted features. Effectiveness of the feature vector in terms of intra-person variation and inter-persons
variation was obtained from feature plots.
Keywords: Center points, Edge detected image, Feature Face-image, Pose invariant.
----------------------------------------------------------------------***------------------------------------------------------------------------
1. INTRODUCTION
Recognition of people is usually done using widely accepted
biometric traits like signature, fingerprint and face. Face
identification can be done naturally by people or artificially
using intelligent machines. Naturally people find it easy to
identify faces that are well-known compare to strange faces.
Machine based face recognition involves capturing of face
images using digital camera under variable facial expression
thereafter captured face images are sent to system for
identification. Face images contain important revealing parts
like forehead, eyes, nose, mouth and chin. These parts occupy
different locations and vary closely from one person to
another. The major challenge that makes automatic face
identification difficult is high intra variation within face
images of the same person. This intra variation is caused
majorly by variable illumination and pose.
Face recognition systems comprise of face detection, face
image preprocessing, feature extraction, training and
matching. Feature extraction subsystem is mainly considered
in this work. At preprocessing level many morphological
processing are done to normalize illumination effect [1]. One
method of reducing effect of pose variation is by using robust
feature as input data to classification algorithm. Extraction of
feature from face images can be done in many ways. In the
past many researchers have used methods that involves all the
pixels of the whole image [2][3]. On many occasions feature
are extracted from vital parts of face images [4][5]. Also face
images may be decomposed into smaller image blocks before
feature extraction is carried out [6][7][8].
Extraction of feature vector was carried out using group of
pixel values within eyes, lip and nose regions in [9]. The
feature vector size was reduced and further processed by
application of Principal Component Analysis (PCA) and
Linear Discriminant Analysis (LDA). In [10], face feature
extraction was done using three multi-scale representation
techniques based on Gabor filter, log Gabor filter and Discrete
Wavelet Transform whereas in [11], invariant face feature was
extracted for recognition purpose using Haar wavelet
transform and Principal Component Analysis .
In this work new approach for feature extraction different
from those used in previous works is introduced. The
proposed feature will suppress effect of varying facial
expression. This is achieved by extracting local shape features
from smaller image blocks. And this feature is fused with
texture feature from detailed bands. The rest of the paper is
organized as follows: Section 2 describes the collection of
input images and decomposition. Section 3 introduces the new
feature extraction method, and section 4 describes feature plot
result. Finally, conclusion is presented in section 5.
2. INPUT FACE IMAGES
Digital camera was use to capture face images of people under
variable illumination and pose conditions. Fig.1 shows set of
face image obtained as the input image to the proposed feature
extraction algorithm.](https://image.slidesharecdn.com/effectivefacefeatureforhumanidentification-140812054042-phpapp01/85/Effective-face-feature-for-human-identification-1-320.jpg)
![IJRET: International Journal of Research in Engineering and Technology eISSN: 2319-1163 | pISSN: 2321-7308
__________________________________________________________________________________________
Volume: 03 Issue: 04 | Apr-2014, Available @ http://www.ijret.org 120
0 2 4 6 8 10 12 14 16
0
10
20
30
40
50
60
70
80
90
100
face1
face2
Fig-8: Inter-variation feature plot.
5. CONCLUSIONS
This work presents extraction of fused feature from face
images. The feature is robust enough to reduce intra-face
images variation and increases inter-face images variation.
This feature is produced from first level decomposition of face
image using wavelet transform. Geometric feature obtained
from approximation band is combined with texture feature
from other bands to generate the fused feature vector. Test
results proves that proposed face feature extraction method is
robust enough to reduce the effect of varying face pose for
effective face recognition
REFERENCES
[1]. J. Shermina and V. Vasudevan, 2011, An Efficient Face
Recognition System Based on the Hybridization of Invariant
Pose and Illumination Process European Journal of Scientific
Research, Vol.64, No.2, pp. 225-243.
[2]. K.. Meethongjan, M. Dzulkifli,. A. Rehman and T. Saba,
2010,.Face Recognition based on Fusion of Voronoi Diagram
Automatic Facial and Wavelet Moment Invariants,
International Journal of Video & Image Processing and
Network Security IJVIPNS (10) 4, pp1-8.
[3]. Sarawat Anam, Md. Shohidul Islam, S. Member, , M. A
Kashem, M.A Rahman, 2009, Real Time Face Recognition
Using Step Error Tolerance BPN International Journal of
Engineering and Technology Vol. 1(1) pp.1793-8236.
[4]. Jang-Seon Ryu and Eung-Tae Kim, 2006, Development of
Face Tracking and Recognition Algorithm for DVR(Digital
Video Recorder), International Journal of Computer Science
and Network Security, VOL.6 No.3A, pp.17-24.
[5]. S. Kumar Paul, M. Shorif Uddin and S.Bouakaz, , 2012 ,
Extraction of Facial Feature Points Using Cumulative
Histogram, International Journal of Computer Issues, Vol.9,
Issue 1, No3, pp.44-51.
[6]. T.Venkat Narayana Rao, V.Subramanya Aditya,
S.Venkateshwarlu, B.Vasavi, 2011, Partition Based Face
Recognition System, Journal of Global Research in Computer
Science Volume 2, No. 9, pp.34-38.
[7]. Raju, U.S.N., A. Srikrishna, V. Vijaya Kumar and A.
Suresh, 2008, .Extraction of Skeleton Primitives on Wavelets,
Journal of Theoretical and Applied Information Technology,
pp.1065 – 1074.
[8]. Azadeh Bastani, Esmaeel Fatemi Behbahani, 2011, An
Efficient Feature Extraction Method with Orthogonal
Moments and Wavelet Transform for Human Face
Recognition System European Journal of Scientific Research,
Vol.52 No.3 pp.313-320.
[9]. Dong-Liang, Lee and Jen-sheng Lang, 2010, A face
Detection and Recognition System based on Rectangular
Feature Orientation. In Proc. of International Conference on
System and Engineering, pp.495-499.
[10]. Murugan, D, Dr. S Arumugam, K Rajalakshmi and T. I
Manish, 2010, Performance evaluation of face recognition
using Gabor filter, log Gabor filter and discrete wavelet
transform. International Journal of Computer Science and
Information Technology (2)1, pp.125-133.
[11]. S. Adebayo Daramola and O.Sandra Odeghe, 2012,
Efficient Face Recognition system using Artificial Neural
Network, International Journal of Computer Applications
Vol.41, pp12-15.](https://image.slidesharecdn.com/effectivefacefeatureforhumanidentification-140812054042-phpapp01/85/Effective-face-feature-for-human-identification-4-320.jpg)
Effective face feature for human identification
- 1. IJRET: International Journal of Research in Engineering and Technology eISSN: 2319-1163 | pISSN: 2321-7308 __________________________________________________________________________________________ Volume: 03 Issue: 04 | Apr-2014, Available @ http://www.ijret.org 117 EFFECTIVE FACE FEATURE FOR HUMAN IDENTIFICATION S. Adebayo Daramola1 , Tiwalade Odu2 , Olujimi Ajayi3 1 Senior Lecturer, Department of Electrical and Information Engineering, Covenant University Ota, Ogun State, Nigeria 2 Assistant Lecturer, Department of Electrical & Information Engineering, Covenant University Ota, Ogun State, Nigeria 3 Post Graduate Student, Department of Electrical & Information Engineering, Covenant University Ota, Ogun State, Nigeria Abstract Face image is one of the most important parts of human body. It is easily use for identification process. People naturally identify one another through face images. Due to increase rate of insecurity in our society, accurate machine based face recognition systems are needed to detect impersonators. Face recognition systems comprise of face detector module, preprocessing unit, feature extraction subsystem and classification stage. Robust feature extraction algorithm plays major role in determining the accuracy of intelligent systems that involves image processing analysis. In this paper, pose invariant feature is extracted from human faces. The proposed feature extraction method involves decomposition of captured face image into four sub-bands using Haar wavelet transform thereafter shape and texture features are extracted from approximation and detailed bands respectively. The pose invariant feature vector is computed by fusing the extracted features. Effectiveness of the feature vector in terms of intra-person variation and inter-persons variation was obtained from feature plots. Keywords: Center points, Edge detected image, Feature Face-image, Pose invariant. ----------------------------------------------------------------------***------------------------------------------------------------------------ 1. INTRODUCTION Recognition of people is usually done using widely accepted biometric traits like signature, fingerprint and face. Face identification can be done naturally by people or artificially using intelligent machines. Naturally people find it easy to identify faces that are well-known compare to strange faces. Machine based face recognition involves capturing of face images using digital camera under variable facial expression thereafter captured face images are sent to system for identification. Face images contain important revealing parts like forehead, eyes, nose, mouth and chin. These parts occupy different locations and vary closely from one person to another. The major challenge that makes automatic face identification difficult is high intra variation within face images of the same person. This intra variation is caused majorly by variable illumination and pose. Face recognition systems comprise of face detection, face image preprocessing, feature extraction, training and matching. Feature extraction subsystem is mainly considered in this work. At preprocessing level many morphological processing are done to normalize illumination effect [1]. One method of reducing effect of pose variation is by using robust feature as input data to classification algorithm. Extraction of feature from face images can be done in many ways. In the past many researchers have used methods that involves all the pixels of the whole image [2][3]. On many occasions feature are extracted from vital parts of face images [4][5]. Also face images may be decomposed into smaller image blocks before feature extraction is carried out [6][7][8]. Extraction of feature vector was carried out using group of pixel values within eyes, lip and nose regions in [9]. The feature vector size was reduced and further processed by application of Principal Component Analysis (PCA) and Linear Discriminant Analysis (LDA). In [10], face feature extraction was done using three multi-scale representation techniques based on Gabor filter, log Gabor filter and Discrete Wavelet Transform whereas in [11], invariant face feature was extracted for recognition purpose using Haar wavelet transform and Principal Component Analysis . In this work new approach for feature extraction different from those used in previous works is introduced. The proposed feature will suppress effect of varying facial expression. This is achieved by extracting local shape features from smaller image blocks. And this feature is fused with texture feature from detailed bands. The rest of the paper is organized as follows: Section 2 describes the collection of input images and decomposition. Section 3 introduces the new feature extraction method, and section 4 describes feature plot result. Finally, conclusion is presented in section 5. 2. INPUT FACE IMAGES Digital camera was use to capture face images of people under variable illumination and pose conditions. Fig.1 shows set of face image obtained as the input image to the proposed feature extraction algorithm.
- 2. IJRET: International Journal of Research in Engineering and Technology eISSN: 2319-1163 | pISSN: 2321-7308 __________________________________________________________________________________________ Volume: 03 Issue: 04 | Apr-2014, Available @ http://www.ijret.org 118 2.1 Face Image Decomposition Firstly, input colour image is resized and converted to grayscale image as shown in fig.2. The image is passed to Haar wavelet transform algorithm. The image is decomposed into four sub-bands (one approximation band and three detailed bands). The output image obtained from this stage is as shown in fig3. Fig-1: Examples of captured face images. Fig-2: Sample of input image. Fig-3: Output of image decomposition. 3. FEATURE EXTRACTION METHOD Feature extraction technique describes in this section produce fused feature that is able to capture pose variation caused by different facial expression. It is well known that facial components like eyes, nose, mouth and chin have high gray level intensity than the surrounding therefore provide distinguishable edge information. Smooth contour of facial components are created by performed edge detection operation on approximation band. Canny edge detection was used for this operation. The output image was resized as shown in Fig.4. 3.1 Contour Face Feature Robust geometric feature is extracted from pixels positions of the edge detected image. The feature is extracted using the following steps. 1. Split the edge detected image into two parts (i) Calculate centre of gravity of the image. (ii) Partition the image vertically into two image blocks through the centre of gravity as shown in Fig.5. Fig-4: Edge detected image. 2. Split each of the image-blocks parts obtained from step one into four smaller image-blocks (i) Calculate centre of gravity of each of the image (ii) Partition each of the images through the centre of gravity into four smaller image blocks as shown in Fig.6.
- 3. IJRET: International Journal of Research in Engineering and Technology eISSN: 2319-1163 | pISSN: 2321-7308 __________________________________________________________________________________________ Volume: 03 Issue: 04 | Apr-2014, Available @ http://www.ijret.org 119 3. Obtain two geometric features (i) Calculate the mean area (mA) of the connected components in each of the eight smaller image blocks as in equation1. (ii) Calculate the mean perimeter (mP) of the connected components in each of the eight smaller image blocks as in equation2. 4. Concatenate the two features in step (i) and (ii) to obtain geometric feature vector as in equation3. Fig-5: Output of vertical splitting. Fig-6: Eight image blocks. 3.2 Textural Face Feature Texture feature is obtained from the detailed bands using Singular Value Decomposition (SVD). SVD of image H is defined as H = WSVT . Matrix W is the m × m dimension matrix of eigenvectors of the covariance matrix HHT and matrix S is an m × n rectangular diagonal matrix whereas matix V is the n × n matrix of eigenvectors of HT H. Given that F1(w,s,v), F2(w,s,v) and F3(w,s,v) are the Single Value Decomposition features extracted from the LH, HL and HH band respectively. Texture feature is calculated using only the first coefficient of the diagonal matrix. Given that F1(s), F2(s) and F3(s) are the coefficient value of diagonal matrix from band LH, HL and HH respectively. Therefore texture feature (T) is calculated as in equation 4. T = (F1(s) + F2(S) + F3(s))/3 (4) 3.3 Fused Face Feature At this stage the features extracted from the approximation band and detail bands are fused together to get the feature that best describe the image. The fuse feature contains relevant information from geometric and texture attribute of the face image. In this work, geometric feature (G) and texture feature (T) are combined at the feature level as it written in equation 5. The fused feature vector obtained at this stage can use as input data to train suitable classifier for face recognition. 4. FEATURE TEST Intra-variation and inter-variation of the proposed feature are experimental by plotting the feature graph of two face images of the same person as shown in Fig.7. Also a feature graph of two face images from different persons is plotted as shown in Fig.8. 0 2 4 6 8 10 12 14 16 0 50 100 150 200 250 face1 face2 Fig-7: Intra-variation feature plot
- 4. IJRET: International Journal of Research in Engineering and Technology eISSN: 2319-1163 | pISSN: 2321-7308 __________________________________________________________________________________________ Volume: 03 Issue: 04 | Apr-2014, Available @ http://www.ijret.org 120 0 2 4 6 8 10 12 14 16 0 10 20 30 40 50 60 70 80 90 100 face1 face2 Fig-8: Inter-variation feature plot. 5. CONCLUSIONS This work presents extraction of fused feature from face images. The feature is robust enough to reduce intra-face images variation and increases inter-face images variation. This feature is produced from first level decomposition of face image using wavelet transform. Geometric feature obtained from approximation band is combined with texture feature from other bands to generate the fused feature vector. Test results proves that proposed face feature extraction method is robust enough to reduce the effect of varying face pose for effective face recognition REFERENCES [1]. J. Shermina and V. Vasudevan, 2011, An Efficient Face Recognition System Based on the Hybridization of Invariant Pose and Illumination Process European Journal of Scientific Research, Vol.64, No.2, pp. 225-243. [2]. K.. Meethongjan, M. Dzulkifli,. A. Rehman and T. Saba, 2010,.Face Recognition based on Fusion of Voronoi Diagram Automatic Facial and Wavelet Moment Invariants, International Journal of Video & Image Processing and Network Security IJVIPNS (10) 4, pp1-8. [3]. Sarawat Anam, Md. Shohidul Islam, S. Member, , M. A Kashem, M.A Rahman, 2009, Real Time Face Recognition Using Step Error Tolerance BPN International Journal of Engineering and Technology Vol. 1(1) pp.1793-8236. [4]. Jang-Seon Ryu and Eung-Tae Kim, 2006, Development of Face Tracking and Recognition Algorithm for DVR(Digital Video Recorder), International Journal of Computer Science and Network Security, VOL.6 No.3A, pp.17-24. [5]. S. Kumar Paul, M. Shorif Uddin and S.Bouakaz, , 2012 , Extraction of Facial Feature Points Using Cumulative Histogram, International Journal of Computer Issues, Vol.9, Issue 1, No3, pp.44-51. [6]. T.Venkat Narayana Rao, V.Subramanya Aditya, S.Venkateshwarlu, B.Vasavi, 2011, Partition Based Face Recognition System, Journal of Global Research in Computer Science Volume 2, No. 9, pp.34-38. [7]. Raju, U.S.N., A. Srikrishna, V. Vijaya Kumar and A. Suresh, 2008, .Extraction of Skeleton Primitives on Wavelets, Journal of Theoretical and Applied Information Technology, pp.1065 – 1074. [8]. Azadeh Bastani, Esmaeel Fatemi Behbahani, 2011, An Efficient Feature Extraction Method with Orthogonal Moments and Wavelet Transform for Human Face Recognition System European Journal of Scientific Research, Vol.52 No.3 pp.313-320. [9]. Dong-Liang, Lee and Jen-sheng Lang, 2010, A face Detection and Recognition System based on Rectangular Feature Orientation. In Proc. of International Conference on System and Engineering, pp.495-499. [10]. Murugan, D, Dr. S Arumugam, K Rajalakshmi and T. I Manish, 2010, Performance evaluation of face recognition using Gabor filter, log Gabor filter and discrete wavelet transform. International Journal of Computer Science and Information Technology (2)1, pp.125-133. [11]. S. Adebayo Daramola and O.Sandra Odeghe, 2012, Efficient Face Recognition system using Artificial Neural Network, International Journal of Computer Applications Vol.41, pp12-15.