Discrete Wavelet Transform in Face Recognition

Transcription

1 Discrete Wavelet Transform in Face Recognition Mrs. Pallavi D.Wadkar 1, Mr. Jitendra M.Bakliwal 2 and Mrs. Megha Wankhade 3 1 Assistant Professor, Department of Electronics and Telecommunication,University of Pune Marathwada Mitra Mandal s Institute of Technology, Lohgaon,, Pune, India 2 Assistant Professor & Head of Department of Electronics and Telecommunication, University of Pune Marathwada Mitra Mandal s Institute of Technology, Lohgaon, Pune, India 3 Assistant Professor, Department of Electronics & Telecommunication, University of Pune Sinhgad College of Engineering, vadgaon(bk), Pune, India. Abstract: Face recognition is emerging as an active research area spanning several disciplines such as image processing, pattern recognition, computer vision and neural networks. This paper presents face recognition using Haar wavelet filter which is type of discrete wavelet transform has been implemented as a part of the proposed algorithm due to its simplicity, suitability and regularity for face recognition using multi resolution approaches. A range of wavelet decompositions have been implemented in order to investigate the best performances. ORL face database is used for experiment purpose. Distance measure such as Euclidean distance is used. Result analysis shows that proposed method is more effective than available other methods in improving verification performance of face recognition system. Keywords: Face recognition, Discrete Wavelet Transform, Haar wavelet, ORL database, Euclidean distance. I. INTRODUCTION In recent years, face recognition has been the subject of intensive research. With the current perceived world security situation, governments as well as businesses require reliable methods to accurately identify individuals, without overly infringing on rights to privacy or requiring significant compliance on the part of the individual being recognized. Face recognition provides an acceptable solution to this problem. Face recognition has drawn attention of the research community. Face identification from a single image is a challenging task because of variable factors like alterations in scale, location, pose, facial expression, occlusion, lighting conditions and overall appearance of the face. With the synergy of efforts from researchers in diverse fields including computer engineering, mathematics, neuroscience and psychophysics, different frameworks have evolved for solving the problem of face recognition All face recognition algorithms consists of two parts: a) Face localization & normalization b) Face identification. Partially automatic algorithms are given a facial image and the coordinates of centre of eyes. Fully automatic algorithms are only given facial images. A multitude of techniques have been applied to face recognition. Earliest methods treated faces as points in very high dimensional space and calculated the Euclidean distance between them. Dimensional reduction techniques including Principal Component Analysis (PCA) [2], [3], [7] have now been successfully applied to the problem, thus reducing complexity of the recognition process without negatively infringing on accuracy. Another technique that has been applied to the field is Neural Network Models (NMM) [4]. With NMM, the system is supplied with a training set of images along with correct classification, thus allowing the neural network to ascertain a weighting system to determine which areas of an image are deemed most important. A further technique Discrete Wavelet Transform (DWT) [5], [6] has also been used within the field of face recognition. The main reasons for its popularity lie in its complete theoretical framework, the great flexibility for choosing bases and the low computational complexity. Wavelet Transform is a signal analysis method of the time scale and has a advantage of multi-resolution analysis. It is the time-frequency localization analysis which has the capacity of local features in the time domain and frequency domain. It has a fixed window, but the shape of its time window and frequency window can change. With wavelets, one can analyse low-frequency components of the signal with high-frequency resolution while at the same time looking at high-frequency components with high time resolution. Because it offers this capability, the wavelet transform has attracted much attention and has enjoyed successful applications in many fields of science and engineering, including pattern recognition, data compression, financial analysis, and measurement science fields. The paper is organized as follows. The proposed system is described in detail in section 2. The experimental results and discussion are given in section 3 while concluding remarks are given in section 4. Volume 2, Issue 3 May June 2013 Page 60

2 II. PROPOSED SYSTEM This paper concerns face recognition using multi resolution analysis, namely wavelet decomposition.. Query Image Wavelet Decomposition Obtain Energie s Data base Images Obtain Energie s Wavelet Decomposition Euclidean Distance Sort Results Fig. 1 Proposed System Figure 1 shows block diagram for the proposed system. The wavelet transform provides a powerful mathematical tool for analysing non-stationary signals. The images used in this paper have been taken from the ORL database. A. Wavelet Transform Wavelet Transform is a popular tool in image processing and computer vision. Many applications, such as compression, detection, recognition, image retrieval et al. have been investigated. WT has the nice features of space frequency localization and multi-resolutions. 1-D continuous WT of function f(t) defined as is wavelet basis function. is called mother wavelet which has at least one vanishing moment. The arguments and denote the scale and location parameters, respectively. The oscillation in the basis functions increases with a decrease in a. The transform can be discretized by restraining and to a discrete lattice. 2-D DWT is generally carried out using a separable approach, by first calculating the 1-D DWT on the rows, and then the 1-D DWT on the columns : DWTn[DWTm[x[m,n]].Two-dimensional WT decomposes an image into 4 subbands that are localized in frequency and orientation, by LL, HL, LH, HH. Each of these sub bands can be thought of as a smaller version of the image representing different image properties. The band LL is a coarser approximation to the original image. The bands LH and HL record the changes of the image along horizontal and vertical directions, respectively. The HH band shows the high frequency component of the image. Fig. 2 Two-level wavelet decompositions of two images Second level decomposition can then be conducted on the LL sub band. Fig.2 shows a two-level wavelet decomposition of two images of size 112X92 pixels. Earlier studies concluded that information in low spatial frequency bands play a dominant role in face recognition. Nastar et al. have investigated the relationship between variations in facial appearance and their deformation spectrums [7]. They found that facial expressions and small occlusions affect the intensity manifold locally. Under frequency-based representation, only highfrequency spectrum is affected, called high-frequency phenomenon. Moreover, changes in pose or scale of a face affect the intensity manifold globally, in which only their lowfrequency spectrum is affected, called low-frequency phenomenon. Only a change in face will affect all frequency components. In their recent work on combining wavelet sub band representations with Eigen face methods, Lai et al. also demonstrated that: 1) the effect of different facial expressions can be attenuated by removing the high-frequency components and 2) the low-frequency components only are sufficient for recognition. So in the following, we will mainly use low-frequency subband coefficients for recognition to attenuate natural difference in the images of the same person. Further decomposition to the LL subband (two-level decomposition), leads to lower dimensionalities and a multi resolution image. We could perform higher levels of decomposition, like three or four level decomposition, even higher. The number of levels we choose depends on our work and need. Fig.6 shows Biquadratic wavelet decomposition. B. Haar Wavelet Transform (HWT) Volume 2, Issue 3 May June 2013 Page 61

3 HWT decomposition works on an averaging and differencing process as follows [5]: end procedure The second type of two-dimensional wavelet transform, called the nonstandard decomposition, alternates between operations on rowsand columns. First, we perform one step of horizontal pairwise averaging and differencing on the pixel values in averaging and differencing to each column of the result. It can be seen that the number of decomposition steps is 22 = 4. Given an original image, the Haar wavelet transform method separates high frequency and low frequency bands of the image by high-pass and low-pass filters from the horizontal direction, and so does the vertical direction of the image. 1) Two-dimensional Haar wavelet transforms: There are two ways we can use wavelets to transform the pixel values within an image. Each is a generalization to two dimensions of the one-dimensional wavelet transform.to obtain the standard decomposition of an image; we first apply the one-dimensional wavelet transform to each row of pixel values. This operation gives us an average value along with detail coefficients for each row. Next, we treat these transformed rows as if they were themselves an image and apply the one-dimensional transform to each column. The resulting values are all detail coefficients except for a single overall average coefficient. The algorithm below computes the standard decomposition. Figure 3 illustrates each step of its operation. Fig. 4 Nonstandard decomposition of an image To complete the transformation, we repeat this process recursively only on the quadrant containing averages in both directions. Figure 4 shows all the steps involved in the nonstandard decomposition. procedure NonstandardDecomposition(C: array [1.. h, 1.. h] of reals) C C=h (normalize input coefficients) while h > 1 do for row 1 to h do DecompositionStep (C[row, 1.. h]) for col 1 to h do DecompositionStep (C[1.. h, col]) h h=2 end while end procedure Fig. 1 Standard decomposition of an image procedure StandardDecomposition (C: array [1.. h, 1.. w] of reals) for row 1 to h do Decomposition (C[row, 1.. w]) for col 1 to w do Decomposition (C[1.. h, col]) C. Euclidean distance method The most common recognition rule is the use of the Euclidean distance. Euclidean distance calculates the shortest Euclidean distance from the query image and database images. Euclidean distance =. i Where F is feature space of database image & Fr is feature space of query image Volume 2, Issue 3 May June 2013 Page 62

4 III. THE SIMULATION EXPERIMENT & RESULT ANALYSIS The ORL standard face database is chosen in this simulation experiment, Simulate on the platform of Matlab7.5. The ORL standard face database consists of 400 face images attained from 40 people. Each people has 10 images of different expression or gesture. The resolution rate of the image is and the gray scale is 256. Work presented in [9] uses artificial neural network approach for face recognition.table III shows that proposed (the method applied in the present study) method had given 0% False Acceptance Rate and False Acceptance Rate as compared to the work mentioned. This implies that proposed method is more effective than available other methods in improving verification performance of face recognition system. Table I: Recognition results using different sets of images per person in database No. of Images Tested No. of Images in database per person Recognition Rate(%) (Euclidean distance) Fig. 5 Parts of images of the ORL facial database From the Table I, it is observed that, when experimentation is done with 10 no. of images per person in database, recognition rate is improved. Table II: Recognition results using different decomposition levels No. of Recognition decomposition Rate(%) levels From the Table II, it is observed that, Recognition rate are improved when no of decompositions are increased. In the reference [8], it also points out that the best results are obtained by the use of images of pixels of 7 x 6 in the experiment about statistical unrelated optimal discriminate transform, which is based on the ORL database. As said above, it shows that recognition results using low-pixel facial images can be better than those using high-pixel facial images. Table III: Performance comparison with another work Method used Work presented in [9] False Acceptance Rate (%) False Rejection Rate (%) Proposed Method Fig. 6 The example of biquadratic wavelet decomposition Volume 2, Issue 3 May June 2013 Page 63

5 IV. CONCLUSIONS The proposed Face Recognition Algorithm using Haar Wavelet Transform is implemented and tested for ORL. Distance measures such as Euclidean measure is used in our work. Recognition results are improved when there are more no of images per person in database. When no of decompositions are increased then also Recognition rate is improved. Future work includes improvement of the performance of the proposed method. Further implementing the proposed algorithm with more multi-resolution transforms and finding the transform which will give maximum recognition rate.the further research-orientation in the future is to design the algorithm which give more accurate face recognition. REFERENCES [1] A. Amira & P. Farrell, An Automatic Face Recognition System Based on Wavelet Transforms Institute of Electronics, Communications and Information Technology (ECIT) 2005 IEEE. [2] M. Turk and A. Pentland Eigenfaces for Recognition, J. Cognitive NeuroScience, vol. 3, pp , [3] B. Moghaddan, W. Wahid and A. Pentland, Beyond Eigenfaces: Probabilistic Matching for Face Recognition, Proceedings of Face and Gesture Recognition, pp 30-35, [4] M. Bicego, U. Castellani and V. Murino Using Hidden Markov Models and Wavelets for facerecognition ICIAP 03, 12th International Conference on Image Analysis and Processing,September 17-19, [5] E. J. Stollnitz, T. D. DeRose and D. H. Salesin" Wavelets for computer graphics: a primer, part I,"IEEE Computer Graphics and Applications, vol.15, No. 3, pp , May [6] M. Harandi, M. Ahmadabadi and B. Araabi FaceRecognition Using Reinforcement Learning Proceedings of ICIP 04, International Conference on Image Processing. [7] C. Nastar and N. Ayach, Frequency-Based Nonrigid Motion Analysis,IEEE Trans. Pattern Anal. Mach. Intell., vol 18: , Nov [8] Zhong Jin, Face recognition based on the uncorrelated discriminant transformation, Pattern Recognition, 2001, Vol. 34(1), pp [9] Shahrin Azuan Nazeer 1, Nazaruddin Omar and Marzuki Khalid, Face Recognition System using Artificial Neural Networks Approach IEEE - ICSCN 2007, MIT Campus, Anna University, Chennai, India.. pp Feb , Volume 2, Issue 3 May June 2013 Page 64