• Tidak ada hasil yang ditemukan

View of ANALYSIS AND PERFORMANCE OF FACE RECOGNITION SYSTEM USING COMPLEX WAVELET WITH HIDDEN MARKOV MODEL

N/A
N/A
Protected

Academic year: 2023

Membagikan "View of ANALYSIS AND PERFORMANCE OF FACE RECOGNITION SYSTEM USING COMPLEX WAVELET WITH HIDDEN MARKOV MODEL"

Copied!
8
0
0

Teks penuh

(1)

1

ANALYSIS AND PERFORMANCE OF FACE RECOGNITION SYSTEM USING COMPLEX WAVELET WITH HIDDEN MARKOV MODEL

ANKITA NIGAM (Research Scholar)

Department of CS, Mewar University Mewar india, DR. SAMAR UPADHYAY(HOD)

Department of MCA,JEC Jabalpur India)

Abstract Face recognition has been studied for several years in the context of biometrics and is one of the most successful applications of image analysis and understanding. This thesis focuses on face identification method, for recognition of individual face image from face database. Face identification method is a means of identifying different faces against several stored pattern faces. It takes image of a human face as input and searches for a match in the stored face images. If there is a match, the user can see the result as the face recognized or not recognized. User cannot generate any kind of modification in the stored image files, i.e. a user is not allowed to insert, delete, update or modify images from the storage data. The administrator of the system has validation to make modifications in the storage data or face image database. The intention of this research is to develop an original, correct and efficient face recognition system. This system uses the wavelet filter to extract the sequence of informative wavelet features from the given facial image. The extracted features are again subjected to form a sequence of feature vectors from an image. Then, the system developed uses the Hidden Markov Model (HMM) to match a test facial image with an accurate reference image. Viterbi algorithm is used to find the highest possibility of the surveillance and analysis order

Keywords:CWT, HMM,DRT, DT-CWT, ORL, YALE, SPACEK, JAFFE, and FERET I. INTRODUCTION1

Face recognition (FR) has emerged as one of the most extensively studied research topics that spans multiple disciplines such as pattern recognition, signal processing and computer vision (W. Zhao et.al.,2003).This is due to its numerous important applications in identity authentication, security access control, intelligent human-computer interaction, and automatic indexing of image and video databases. Face recognition has repeatedly shown its importance over the last ten years or so. Not only is it a vividly researched area of image analysis, pattern recognition and more precisely biometrics but also it has become an important part of our everyday lives since it was introduced as one of the identification methods to be used in e- passports and in many security purposes. The human face is full of information but working with all the information is time consuming and less efficient. It is better get unique and important information and discards other

1

useless information in order to make system efficient. Face recognition (W.

Zhao et.al.,2003) systems can be widely used in areas where more security is needed. Researchers have developed varieties of new techniques to improve the face recognition rate. Also different people gone for different approaches like Geometric /Template Based Approaches,

Piecemeal/ Holistic

Approaches,Template/Statistical/ Neural Network Approaches (D. Rana and Dr. N.

P. Rath,2012)

In this paper we have done A biometrics method for performing identification, of automatic face recognition is presented of digital image.

In this method the images from a database are of different size and format, are to be converted into the standard dimension which is appropriate for applying DT-CWT to face images to provide a better representation for feature extraction data is based on wavelet features extraction using wavelet filter bank. The input image reduces image dimension by decomposition with wavelet filter for feature extraction and generation algorithm is used to choose a set of useful and non-redundant wavelet features. The DT-CWT is applied on LL

(2)

sub band, to generate DT-CWT coefficients to form feature vectors. HMM (Hidden Markov Models) is used for matching the input face image to the template images.

II Implementation of Proposed Face Recognition Algorithm Method In our proposed method, a face image is modeled by the following steps:

1. Image is preprocessed:

A two-dimensional function, of an image may be defined as f (x, y), where x and y are spatial (plane) coordinate, and the amplitude of f at any pair of coordinates (x, y) is termed the intensity or gray level of the image at that point. When x, y, and the intensity values of f are all finite, distinct quantities, we call the image a digital image. The sector of digital image processing refers to processing digital images by means of a digital computer.

Note that a digital image is consists of a finite number of elements, each of which has a particular location and worth.

These elements are called picture elements, image elements, pels, and pixels. Pixel is the term used most broadly to indicate the basics of a digital image (Gonzalez et al., 2006).The face images in the database and test image are preprocessed before extracting the facial features. In pre-processing (i) Color to grayscale image conversion; with intensity values between 0 and 255 of the grayscale image is attained from the color image to decrease processing time (ii) Image cropping; face image has a background and other occlusions that may not be necessary to recognize a person correctly fig.1.

(a) (b)

Figure 1: (a) Original image (b) Cropped image.

2. After the original size of Face images are re-sized to the required sizes then a face image is transformed into complex wavelet domain by a 4-level dual-tree complex transform. The face image is decomposed into 4 levels and 6(4) sub bands are obtained.

3. Each labeled sub band is divided into multiple non-overlapping rectangle sub regions.

4. For each cell, the directional derivatives along ±15°, ±45°, and ±75° are computed, Figure 2 show two- dimensional dual-tree complex wavelets in spatial domain and their (idealized) support of the spectrum in 2-D frequency plane on these orientations.

Figure 2: Two-dimensional dual-tree complex wavelets in spatial domain and their (idealized) support of the spectrum in 2-D frequency plane, oriented at

±15°, ±45°, and ±75°, respectively (Selesnick et al., 2005).

5. After extracting CWT features.

Modelling of image feature vector is done with HMM.

6. Euclidean distance between feature vectors of images in database and feature vectors of test image is computed.

7. Image with minimum Euclidean distance is considered as matched image.

Fig. 3 shows the frequency-domain partitions of the DT-CWT, the DT-CWT performs has nearly no overlapping and with full coverage in the frequency- domain representation. Therefore, the feature vectors extracted from the 2D DT- CWT representations should process a better frequency characteristics description for face images. First, each face image is decomposed into multiple sub bands using the 2D DT-CWT representation with 6 different angles and 4 scales (thus, 24 sub bands). A feature vector is constructed that consists of all the 24 sub bands.

(3)

3 Figure 3: Frequency-domain partition of two directional filters (scales = 2;

directions = 6) DT-CWT.

2.2. Feature Extraction by DT-CWT Feature extraction: In which, various facial features extracting techniques are used by earlier researchers like Edge detection techniques, Principal Component Analysis technique, Discrete Cosine Transform coefficients, Discrete Wavelet Transform coefficients, Complex Wavelet Transform, Dual Tree Complex Wavelet Transform, fusion of different features may be used. In this thesis we are applying DT-CWT.

The face features were extracted from the multi-scale and multi-direction DT-CWT represented domain. To construct the feature vector, the magnitudes of the frequency response from each high-pass sub band are utilized, since each high-pass sub band contains specific directional information.

Note that the low-pass sub band was not used since it’s sensitive to the change of illumination. Let Ou, v(z) indicate the magnitudes of the DT-CWT response of a training face image, where u and v are the scale and the orientation parameters for the DT-CWT representation, respectively (here, u {1, . . . , 4}, v{±15°,•±45°,•±75°}).

Each 2D sub band was re- arranged into a column vector by concatenating the columns (or rows) of the DT-CWT coefficient matrix, and the column vector is then normalized to zero mean and unit variance. Let Ou, v denote a normalized column vector of one DTCWT sub band, the feature vector X of one face image is established by concatenating all the 24 sub bands together as

X = (OT1, +15◦, OT1, +45◦, . . . , OT6,−75)T , where T denotes the transpose operator.

The DT-CWT feature vector X resides in a space of very high dimensionality; that is, X Є RN, where N is the dimensionality of the vector space, which is a fairly large integer. To greatly reduce the feature vector X’s dimension.

The DT-CWT offers extra information with respect to the particular directions such as ±150, ±450, ±750. Unlike the Gabor wavelets, however, the directions of the DT-CWT are fixed, while the sub bands on the Gabor wavelets can be computed in any direction. However, the 2D DT- CWT representation, theoretically speaking, has no overlapping and missing frequency-domain coverage, while the Gabor wavelets representation has both. Thus, the DT-CWT has the potential to achieve improved performance on face recognition (Chao 2009).

The amount of features and size are decrease as the amount of Dual Tree- Complex Wavelet Transform level raises fig 4 shows four-level wavelet decomposition. Decrease in number of features clearly decreases memory requirement and computational time. The amount of features for level-1, level-2, level-3 and level-4 are 393216, 98304, 24576 and 6144 correspondingly, and consequently decrease in image dimension is as depicted in the fig. 5.The DT-CWT features extracted from the training data are input to the HMM for the matching of facial images. (Ramesha et al., 2011)

H-high frequency band L-low frequency band Figure4: Four-level wavelet

decomposition.

(4)

Figure 5: DT-CWT images at different levels.

2.3 Modelling of Image by HMM

Matching: After extraction of features from DT-CWT matching is performed on face database. In which Euclidean Distance (ED), Hamming Distance, Support Vector Machine (SVM), Neural Network (Ranawade 2010) and Random Forest (RF) (Montillo et al., 2009; Peng et al., 2008)may be used for matching. Here we are using HMM technique for the purpose of matching the facial images from the face database (eg- ORL, Yale, FERET, L-Spacek, and JAFFE).

Face Database

The significant section of any face recognition system for personal identification is the data collection and the creation of database. In our research Performance of the proposed face recognition scheme has been tested on standard face databases. We have taken Input: Face Database from L- Spacek (Libor Spacek), Yale Face Database, ORL (Olivetti Research Laboratory), FERET (Facial Recognition Technology), and JAFFE (Japanese Female Facial Expression).

Preprocessing: for Preprocessing read Face image from these training databases, then Test Face Images as shown in fig 6.

Figure 6: Sample face images from the Yale Face Database B.

Firstly, with ORL database, the simulation experiments conducted in this research exploit the Cambridge ORL (Olivetti Research Laboratory) face database (Samaria et al., 1994) consists of 400 images of 40 individuals; out of which 10 images have been taken in frontal poses. Each image has a spatial resolution of 112×92. To facilitate the DT- CWT representation in a unified way, each face image had been re-scaled to 128 × 128 pixels. 2D DT-CWT was performed for subband decomposition of the scaled images. Sample images of the database shown in Fig.7.

Figure 7: Faces of 10 person from ORL Face Database

Secondly, with Yale Face Database B, for each topic in a particular pose, an image with ambient (background) illumination was also captured. The folder enclose 5760 single light source images of 10 subjects each seen in 576 viewing conditions (9 poses x 64 illumination conditions). Therefore, the overall number of images is in fact 5760+90=5850. The total size of the compressed database is about 1GB. From which we have selected 10 images for our research

Figure 8: Face images from Yale Face Database

Thirdly, the JAFFE (Japanese Female Facial Expression) images database contains 213 images posed by 10 Japanese female models. We have

(5)

5 selected 10 images for our research purpose as depicted below:

Fig. 9: Facial Images from JAFFE database

Fourthly, In L- Spacek (Libor Spacek) database total number of individuals are 395, number of images per individual is 20, total number of images is 7900, race contains images of people of various racial origins, age range in the images are

mainly of first- year

undergraduate students, so the majority of individuals are between 18-20 years old but some older individuals are also present and some have glasses. The sample images which we have taken are as below:

Figure 10: Facial Images from L-Spacek database.

Fifthly, we have conducted experiments on commonly used face databases:

FERET (Face Recognition Technology) database (Philipps et al., 2000). For FERET database, 600 frontal face images from 200 subjects 10 images are selected, where all the subjects are in an upright, frontal position. The 600 face images were acquired under varying illumination conditions and facial expressions. Each subject has three images of size 256 × 384 with 256 gray levels. The following procedures were applied to normalize the face images prior to the experiments:

(i) each face image is cropped to the size of 128 × 128 to extract the facial region using the algorithm in ( Nilsson et al., 2007)

(ii) each face image is normalized to zero mean and unit variance.

To test the algorithms, ten images are randomly chosen for training, and then for testing. Figure 11 shows sample images from the FERET database. Figure 12 shows the impulse responses of the dual-tree complex wavelets. It is evident that the transform is selective in 6 directions in all of the scales. Figures 13 show the magnitude and real part of a face image processed using the DT-CWT.

Figure 11: Example FERET images used in our experiments with 10 different faces(cropped to the size of 128 × 128 to extract the facial region).

(a)

(b)

Figure 12: Impulse response of DT-CWT at four levels and six orientations. (a)

Real part. (b) Magnitude.

.

(6)

(a)

(b)

Figure 13: Dual Tree–Complex Wavelet Transformation of a test image (upper left face in Figure 4.8). (a) The magnitude of the transformation. (b) The real part of the transformation Step used perform operation to given data

(a) Re-sizing the image

The size of an image is resized to required size or a Region of Interest (ROI) of an image.

(b)Feature Extraction

The first step in any face recognition system is the extraction of the feature matrix Feature extraction is the process by which the key features of the samples are selected. The process of feature extraction is depending on the set of

algorithms. 4-level DT-CWT on LL subband to generate feature has been used.

(3)Output: Match/Mismatch Face Image by using Hidden Markov Model (HMM).

This method step perform using matlab coding

Result

Graphs for validation and testing time taken for detection of image to analyse the performance ratio are shown below:

Figure 14: Graph2 showing performance and epoch ratio at 20

Figure 15: Graph2 showing performance and epoch ratio at 100

(7)

7 Figure 16: Graph3 showing performance and epoch ratio at 160

Figure 17: Graph4 showing performance and epoch ratio at 200.

In Figure 14 Graph 1 shows ratio between performance and epoch (20), in fig. 15 Graph 2: shows ratio between performance and epoch (100), in fig.16 Graph 3: shows ratio between performance and epoch (160) and in fig.

17 Graph 4 shows ratio between performance and epoch (200) which is closer to aim

Comparison of Recognition Rate of the Proposed Algorithm with Other Algorithms

Databa se Of face

Algorithms DT-

CWT [Peng et al.,20 08;Pri ya etal., 2010;

Yue- Hui Sun et al., 2006]

ADT - CWP [Yue -Hui Sun et al., 200 6]

Blo cke d Ba sed

DT - CW T(μ , σ)[

Pri ya et al.,

20 10]

Loca l Fusi

on DT- CWT [Priy a et al., 2010 ]

DDF RMC WTH MM

ORL 76.6

__ 78.

4 82.2 99 YALE 88.6 95.3 90.

3 93.3 98 L-

SPACE K

__ __ __ __ 96

JAFEE __ __ __ __ 98 FERET __ __ __ __ 97 We have used, standard face databases (ORL, YALE, L-SPACEK, JAFFE, and FERET) on which extensive experimentation is carried out and by the proposed method in comparison to those obtained by some of the existing methods a very high degree of recognition accuracy is achieved. This system will be more efficient in comparison to existing work because they have not used CWT along with HMM model to make a sequence of states for matching features of a face. Different illumination of the scene; changes in pose, orientation, expression and face occlusions are some examples of the issues which are deal with earlier researchers.

Conclusion

In this thesis Designing and development of face recognition model using complex wavelet transform with Hidden Markov Model. The ORL, Yale-B, L-Spacek, JAFEE, and FERET database images are used to test the proposed algorithm. DT- CWT is applied to facial images, which

(8)

significantly reduces image dimension by retaining visually significant components of an image and it gives out feature vectors of a face image. Using feature vectors face image is verified using HMM.

The HMM-based system developed in this study matches the feature set (observation sequence) for the test image with the HMM of a claimed image, through viterbi alignment. It is observed that the performance parameters are improved in the case of proposed algorithm compared to the existing algorithms.

Reference

1 Albert Montillo and Haibin Ling, “Age Regression from Faces using Random Forests,” Proceedings of the IEEE International Conference on Image Processing, 2009, pp. 2465-2468

2 Chao-Chun Liu and Dao-Qing Dai “Face Recognition Using Dual-Tree Complex Wavelet Features”, IEEE, 2009.

3 D. Rana and Dr. N. P. Rath, “Face Identification using Soft Computing Tool “, IEEE International Conference, Conference on Advanced Communication Control and Computing Technologies (ICACCCT-2012), Tamilnadu, INDIA, Aug 23-25. Conference Proceeding Page(s): 232- 236, 2012

4 Gonzalez, Rafael C., Richard E. Woods and Steven L. Eddins “Digital Image Processing Using Matlab”

New Delhi (2006).

5 K Jaya Priya and R S Rajesh, “Dual Tree Complex Wavelet Transform based Face Recognition with Single View,” Proceedings of the International Conference on Computing, Communications and Information Technology Applications, 2010, vol. 5, pp. 455-459

6 K Jaya Priya and R S Rajesh, “Local Fusion of Complex Dual-Tree Wavelet Coefficients based Face Recognition for Single Sample Problem,”

International Journal by Elsevier Ltd., 2010,vol. 2, pp. 94-100.

7 M. Nilsson, J. Nordberg, and I. Claesson, “Face detection using local SMQT features and split up snow classifier,” in Proceedings of the IEEE

International Conference on

Acoustics,SpeechandSignalProcessing(ICASSP’07), vol.2,pp.589–592, Honolulu, Hawaii, USA, April 2007

8 P. J. Philipps, H. Moon, S. Rivzi, and P. Ross, “The FERET evaluation methodology fro face-recognition algorithms,” IEEE Transaction on Pattern Analysis and Machine Intelligence, vol. 22, no. 10, pp.

1090–1100, 2000

9 Ramesha K and Raja K B, “Performance Evaluation of Face Recognition based on DWT and DT-CWT using Multi-matching Classifiers,” IEEE International Conference on Computational Intelligence and Communication Systems, pp. 601- 606, 2011.

10 S S Ranawade, “Face Recognition and Verification using Artificial Neural Network,” International

Journal of Computer Applications, 2010, vol.1, no.

14, pp. 21-25

11 Samaria, F. and Young, S. (1994 October). HMM- based architecture for face identification. Image and Vision Computing, vol. 12, pp. 537–543

12 The Yale database

(https://computervisiononline.com/dataset/1105 138686)

13 The JAFFE database

(http://www.kasrl.org/jaffe_info.html)

14 The Libor Spacek database

(http://cswww.essex.ac.uk/mv/allfaces/index.h tml)

15 The FERET

database(https://www.nist.gov/sites/default/files /documents/2016/12/15/feret3.pdf

16 Yue-Hui Sun and Ming-Hui Du, “Face Detection using DT-CWT on Spectral Histogram,”

Proceedings of the IEEE International Conference on Machine Learning and Cybernetics, 2006, pp. 3637 – 3642.

17 Yigang Peng, Xudong Xie , Wenli Xu and Qionghai Dai “Face recognition using anisotropic dual-tree complex wavelet packets “ Pattern Recognition,. ICPR 19th International Conference on IEEE 2008, pp. 1-4.

18 W. Zhao, R. Chellappa, P. J. Phillips, and A.

Rosenfeld, “Face Recognition: A Literature Survey,”

ACM Computing Surveys, vol. 35, no. 4, pp. 399- 458, 2003

Referensi

Dokumen terkait

Make Fashion Patterns Students in groups are asked to recall one size of group of friends, then learners are asked to create a pattern of clothing with the actual size, which in the

HISTORY AND BACKGROUND OF FACE-TO-FACE EDUCATION Face to face education is a teaching methodology where the teacher and students interact with each other by being physically present in