Fusion of Wavelet Coefficients from Visual and Thermal Face Images for Human Face Recognition – A Comparative Study

M. K. Bhowmik, D. Bhattacharjee, M. Nasipuri, D. K. Basu & M. Kundu
International Journal of Image Processing (IJIP), Volume (4): Issue (1) 12
Fusion of Wavelet Coefficients from Visual and Thermal Face
Images for Human Face Recognition – A Comparative Study
Mrinal Kanti Bhowmik mkb_cse@yahoo.co.in
Lecturer, Department of Computer Science
and Engineering Tripura University
(A Central University) Suryamaninagar – 799130,
Tripura, India
Debotosh Bhattacharjee debotosh@indiatimes.com
Reader, Department of Computer Science
and Engineering Jadavpur University
Kolkata – 700032, India
Mita Nasipuri mitanasipuri@gmail.com
Professor, Department of Computer Science
Dipak Kumar Basu dipakkbasu@gmail.com
Professor, AICTE Emeritus Fellow,
Department of Computer Science
Mahantapas Kundu mkundu@cse.jdvu.ac.in
Professor, Department of Computer Science
Abstract
In this paper we present a comparative study on fusion of visual and thermal
images using different wavelet transformations. Here, coefficients of discrete
wavelet transforms from both visual and thermal images are computed
separately and combined. Next, inverse discrete wavelet transformation is taken
in order to obtain fused face image. Both Haar and Daubechies (db2) wavelet
transforms have been used to compare recognition results. For experiments IRIS
Thermal/Visual Face Database was used. Experimental results using Haar and
Daubechies wavelets show that the performance of the approach presented here
achieves maximum success rate of 100% in many cases.
Keywords: Thermal image, Daubechies Wavelet Transform, Haar Wavelet Transform, Fusion, Principal
Component Analysis (PCA), Multi-layer Perceptron, Face Recognition.

1. INTRODUCTION
Research activities on human face recognition are growing rapidly and researchers have
expanded the application potential of the technology in various domains [1, 2, 3, 4, 5]. It is still a
very challenging task to perform efficient face recognition in the outside environments with no
control on illumination [6]. Accuracy of the face recognition system degrades quickly when the
lighting is dim or when it does not uniformly illuminate the face [7]. Light reflected from human
faces also varies depending on the skin color of people from different ethnic groups. Fusion
techniques have many advantages over the thermal and visual images to improve the overall
recognition accuracy. As a matter of fact fusion of images has already established its importance
in case of image analysis, recognition, and classification. Different fusion levels are being used,
for example, low-level fusion, high-level fusion etc. Low-level data fusion combines the data from
two different images of same pose and produces a new data that contains more details. High-
level decision fusion combines the decisions from multiple classification modules [8], [9]. Decision
fusion can be accomplished with majority voting, ranked-list combination [10], and the use of
Dempster-Shafer theory [11]. Several fusion methods have been attempted in face recognition.
Few of them are shown at Table I.
TABLE 1: Different Fusion Methods.
One of the fusion technique mentioned above, A. Gyaourova et al [13] tried to implement pixel-
based fusion at multiple resolution levels. It allows features with different spatial extend to be
fused at the resolution at which they are most salient. In this way, important features appearing at
lower resolutions can be preserved in the fusion process. This method contains two major steps:
one is fusion of IR and visible images and other one is recognition based on the fused images.
Fusion is performed by combining the coefficient of Haar wavelet by decomposition of a pair of IR
and visible images having equal size. They used Genetic Algorithms (GAs) to fuse the wavelet
coefficients from the two spectra. They use GAs for fusion was based on several factors. First,
the search space for the image fusion. Second, the problem at hand appears to have many
suboptimal solutions. They shown the effectiveness of the fusion solution by GAs and they
perform the recognition using eigenfaces. They used Euclidian distance to recognize the PCA
applied by projected data. A. Selinger et al [14] they have compared different algorithms like
eigenfaces (PCA), Linear Discriminate Analysis (LDA) and many more for multiple appearance-
based face recognition methodology. All the experimental data were captured with a visible CCD
array and LWIR microbolometer and used n-fold cross validation technique for the performance.
I. Pavlidis et al [6], they propose a novel imaging solution which address the problems like
lightning and disguise effect in facial image. They demonstrate a dual-band fusion system in the
near infrared which segment human faces much more accurately than traditional visible band
face detection systems and they have also represent with theoretical and experimental
arguments that the upper band of the near infrared is particularly advantageous for disguise
detection purpose. In their face detection system is the only dual band system in the near infrared
range. For this method they calls two cameras one for upper-band (within a range of 0.8 - 1.4 µm)
and lower-band (within a range of 1.4 - 2.2µm). They generated fused image by using weighted
difference method of the co-registered imaging signals from the lower and upper-band cameras
because of the abrupt change in the reflectance for the human skin around 1.4 µm the fusion has
as a result the intensification of the humans face silhouettes and other exposed skin parts and the
diminution of the background. They used Otsu thresholding algorithm for perfect image
Author Technique References
I. Pavlidis Dual band Fusion System [6]
I. Pavlidis Near Infrared Fusion Scheme [12]
J. Heo Data and Decision Fusion [11]
A. Gyaourova Pixel based fusion using Wavelet [13]
A. Selinger Appearance based Face Recognition [14]
A. Selinger Fusion of co-registered visible and LWIR [14]
M. Hanif Optimized Visual and Thermal Image Fusion [15]

segmentation. The dual-band near-infrared face detection method could be used in combination
with a traditional visible spectrum face recognition method. This approach will provide a much
more accurate face segmentation result than a traditional visible band method. G. Bebis et al [16]
in this paper they have extended their previous technique which is mentioned at [13]. In this
paper, they have introduced another fusion scheme as feature-based fusion which is operates in
the eigenspace domain. In this case also they employ a simple and general framework based on
Genetic Algorithms (GAs) to find an optimum fusion strategy and for this feature level fusion they
use eigenspace domain, which involves combining the eigenfeatures from the visible and IR
images. First they compute two eigenspaces, of visible IR face images. They generate the fused
image by applying the GAs to the eigenfeatures from the IR and visible eigenspace. In another
scheme, M. Hanif et al [15] proposed data fusion of visual and thermal images using Gabor
filtering technique. It has been found that by using the proposed fusion technique Gabor filter can
recognize face even with variable expressions and light intensities, but not in extreme condition.
They have designed the data fusion technique in the spatial and DWT domain. For the spatial
domain they have used a weight to generate the fusion image. But in case of DWT image fusion,
they first decompose both thermal and visual image up to level-n and then choose the absolute
maximum at every pixel for both the thermal and visual image to generate the fusion image and
this work is implemented using equinox face database. Jingu Heo et al [11] describe two types of
fusion method as Data fusion and Decision Fusion. In data fusion, it produces an illumination-
invariant face images by integrating the visual and thermal images. A weighted average
technique is used for data fusion. In the data fusion they first detect the eyeglasses and remove
the eyeglasses by a fitting an eclipse and then merge it with visual image. They used n non-
iterative eclipse fitting algorithm for it. They have implemented Decision fusion with highest
matching score (Fh) and Decision fusion with average matching score (Fa). Comparison results
on three fusion-based face recognition techniques showed that fusion-based face recognition
techniques outperformed individual visual and thermal face recognizers under illumination
variations and facial expressions. From them Decision fusion with average matching score
consistently demonstrated superior recognition accuracies as per their results. The organization
of the rest of this paper is as follows: in section 2, the overview of the system is discussed, in
section 3 experimental results and discussions are given. Finally, section 4 concludes this work.
2. SYSTEM OVERVIEW
The block diagram of the system is given in Figure 1. All the processing steps used in this work
are shown in the block diagram. In the first step, decomposition of both the thermal and visual
images up to level five has been done using Haar and Daubechies wavelet transformations. Then
fused images are generated using respective decomposed visual and thermal images of two
different wavelets. These transformed images separated into two groups namely, training set and
testing set are fed into a multilayer perceptron for classification. Before, feeding them into the
network, those are passed through Principal Component Analysis (PCA) for dimensionality
reduction.
2A. Wavelet Transforms and Fusion
Haar wavelet is recognized as the first known wavelet. It is same as Daubechies wavelet (db1).
The Haar wavelet is proposed by Alfred Haar [17]. Haar wavelet is a certain sequence of function.
The Haar wavelet’s mother wavelet function can be described as
(1)
and its scaling function (t) can be described as

(2)
FIGURE 1: System Overview.
It is an orthogonal and bi-orthogonal wavelet. The technical disadvantage of Haar wavelet is that
it is not continuous, and therefore not differentiable. In functional analysis, the Haar systems
denotes the set of Haar wavelets
(3)
The Haar system (with the natural ordering) is further a Schauder basis for the space Lp[0, 1] for
1 ≤ p < +∞. This basis is unconditional for p > 1 [18], [19].
The Haar transform is the simplest of the wavelet transforms. This transform cross-multiplies a
function against the Haar wavelet with various shifts and stretches. The Haar transform is derived
from the Haar matrix. An example of 3x3 Haar matrixes is shown below:

(4)
The Haar transform can be thought of as a sampling process in which rows of the transform
matrix act as samples of finer and finer resolution.
Daubechies wavelets are a family of orthogonal wavelets defining a discrete wavelet transform
and characterized by a maximal number of vanishing moments for some given support. In two
dimensions, the difference of information between the approximations and is
therefore characterized by the sequences of inner products
(5)
(6)
(7)
Each of these sequences of inner products can be considered as an image. In Figure 2(a)
showing level 2 decomposition, H2 gives the vertical higher frequencies (horizontal edges), V2
gives the horizontal higher frequencies (vertical edges), and D2 gives higher frequencies in both
directions (corners). Let us suppose that initially we have an image A1 f measured at the
resolution 1. For any J > 0, this discrete image can be decomposed between the resolution 1 and
2-j
, and completely represented by the 3J + 1 discrete images
(8)
This set of images is called an orthogonal wavelet representation in two dimensions [20]. The
image is a coarse approximation, and the images H2 give the image details for different
orientations and resolutions. If the original image has N
2
pixels, each image , H2, V2, D2 has
2
j
* N
2
pixels (j < 0). The total number of pixels of an orthogonal wavelet represents is therefore
equal to N
2
. It does not increase the volume of data. This is due to the orthogonality of the
representation. Such decompositions were done through Matlab and shown in equations 9 and
10.
[cA, cH, cV, cD] = dwt2 (X, ‘wname’) (9)
[cA, cH, cV, cD] = dwt2 (X, Lo_D, Hi_D) (10)
Equation (9), ‘wname’ is the name of the wavelet used for decomposition. We used ‘haar’ and
‘db2’ as wavelet. Equation (10) Lo_D (decomposition low-pass filter) and Hi_D (decomposition
high-pass filter) wavelet decomposition filters [21], [22], [23], [35], [39].
The more the decomposition scheme is being repeated, the more the approximation of images
concentrates in the low frequency energy. During the decomposition process it actually down-
sampling the rows and columns of an image. Firstly it down-samples the rows (keep one out of
two) and then down-samples the columns (keeps one out of two). The downsampling process for
both the images is same. After decomposing the thermal and visual image we have to generate
the decomposed fused image by providing fusion methods. Then, at each level select the
absolute ‘maximum’ at every pixel from both the images for approximate co-efficient (cA) and
absolute ‘minimum’ at every pixel from both the images for three details co-efficient (cH, cV, cD).

Let us take an example, if T is the thermal and V is the visual image with same pose and
illumination than
‘max’: D = abs(T) ≥ abs(V); C = T(D) + V(~D) (11)
‘min’: D = abs(T) ≤ abs(V); C = T(D) + V(~D) (12)
To generate the fused image (C) of co-efficient it will add that value which is deducted from the
visual image during the calculation of absolute value (D) of thermal (T) and visual (V) image for all
the coefficients using fusion method shown at equation (11) & (12).
Consequently the reconstruction process is performed using inverse of wavelet transformations to
generate synthesized fused image.
X =idwt2 (cA, cH, cV, cD, ‘wname’) (13)
X = idwt2 (cA, cH, cV, cD, Lo_R, Hi_R) (14)
IDWT uses the wavelet ‘wname’ to compute the single-level reconstruction of an Image X, based
on approximation matrix (cA) and detailed matrices cH, cV and cD (horizontal, vertical and
diagonal respectively). By the equation no (14), we can reconstruct the image using filters Lo_R
(reconstruct low-pass) and Hi_R (reconstruct high-pass) and ‘haar’ and ‘db2’ as the time of
reconstruction of an image [33].
FIGURE 2: (a) Shown one of the third level decomposition of level 5 (b) Original image used in the
decomposition (c) Orthogonal wavelet representation of a sample image.
The decomposition process can be iterated, with successive approximations being decomposed
in turn, so that one image is broken down into many lower resolution components. This is called
the wavelet decomposition tree. In this work, decomposition was done up to level five using Haar
and Daubechies (db2) wavelet transformations, as shown in Figure 3. Average of corresponding
transform coefficients from visual and thermal images gives the matrix of fused coefficients, which
when transformed into the image in spatial domain by inverse wavelet transform produces fused
face image. These fused images thus found are passed through PCA for dimensionality
reduction, which is described next.
A3 H3
H2
H1
V3 D3
V2 D2
V1 D1
(a) (b) (c)

FIGURE 3: Wavelet Decomposition Tree.
2B. Dimensionality Reduction
Principal component analysis (PCA) is based on the second-order statistics of the input image,
which tries to attain an optimal representation that minimizes the reconstruction error in a least-
squares sense. Eigenvectors of the covariance matrix of the face images constitute the
eigenfaces. The dimensionality of the face feature space is reduced by selecting only the
eigenvectors possessing significantly large eigenvalues. Once the new face space is constructed,
when a test image arrives, it is projected onto this face space to yield the feature vector—the
representation coefficients in the constructed face space. The classifier decides for the identity of
the individual, according to a similarity score between the test image’s feature vector and the
PCA feature vectors of the individuals in the database [27], [30], [36], [37], [34].
2C. Artificial Neural Network Using Backpropagation With Momentum
Neural networks, with their remarkable ability to derive meaning from complicated or imprecise
data, can be used to extract patterns and detect trends that are too complex to be noticed by
either humans or other computer techniques. A trained neural network can be thought of as an
“expert” in the category of information it has been given to analyze. The Back propagation
learning algorithm is one of the most historical developments in Neural Networks. It has
reawakened the scientific and engineering community to the modeling and processing of many
quantitative phenomena. This learning algorithm is applied to multilayer feed forward networks
consisting of processing elements with continuous differentiable activation functions. Such
networks associated with the back propagation learning algorithm are also called back
propagation networks [24], [25], [26], [27], [28], [29], [38].
3. EXPERIMENTAL RESULTS AND DISCUSSIONS
This work has been simulated using MATLAB 7 in a machine of the configuration 2.13 GHz Intel
Xeon Quad Core Processor and 16384.00 MB of Physical Memory. We analyze the performance
of our algorithm using the IRIS thermal / visual face database. In this database, all the thermal
and visible unregistered face images are taken under variable illuminations, expressions, and
poses. The actual size of the images is 320 x 240 pixels (for both visual and thermal). Total 30
classes are present in that database [31]. Some thermal and visual images and their
corresponding fused images for Haar and Daubechies (db2) wavelets are shown in Figure 4 and
Figure 5 respectively.
To compare results of Haar and Daubechies (db2) wavelets, fusion of visual and thermal images
were done separately. For that purpose 200 thermal and 200 visual images were considered. We
have increased the size of testing image for both the wavelet. Firstly we have tested our system
(341)
(342) (343)
(344)
(6) (7) (8)
(1)
(2) (3) (4)
(0)
(22) (23)(24)
(5)
(86) (87)(88)
(21)
(85)

using 100 fused images using two different wavelet and we increase 10 images per class for 10
different classes i.e. total 100 image. The database is categorized into 10 classes based on
changes in illumination and expressions. The classes with illuminations and expression changes
are class–1, class–2, class–3, class–4, class–6, class–7, and class–9, whereas class–5, class–8
and class–10 are with changes in expressions only. Both the techniques were applied and
experimental results thus obtained are shown in Figure 6.
After increasing the number of testing images the recognition rates are increased for both the
wavelets. Previously we have average recognition rate is 85% for Haar wavelet and 90.019% for
Daubechies wavelet when the number of testing image is 100 (i.e. 10 image per class). Now we
have average recognition rate is 87% for Haar and 91.5% for Daubechies wavelet after increase
the number of testing image 100 to 200 (i.e. 20 image per class).
FIGURE 4: Sample images of IRIS database (a) Thermal images (b) Visual images (c) Corresponding
Fused images using Haar Wavelet transform.
FIGURE 5: Sample images of IRIS database (a) Thermal images (b) Visual images (c) Corresponding
Fused images using Daubechies Wavelet transform.
(a)
(b)
(c)
(a)
(b)
(c)

FIGURE 6: Recognition rate of Haar and Daubechies Wavelet.
In the Figure 6, all the recognition rates for Haar as well as Daubechies wavelet transformations
have been shown. In this figure “dotted” line indicates results using Haar wavelet (with round and
triangle shape marker) and “solid” line indicates results using Daubechies Wavelet (with box and
diamond shape marker). From this figure, it can be easily inferred that overall results using
“Daubechies” (db2) wavelet produces better result than “Haar” wavelet.
Final results have been computed as average of all the categories and given along with results of
other recent methods in Table II. Here, we find results using Daubechies wavelets (db2) shows
comparable results.
TABLE 2: Comparison between different Fusion Techniques.
4. CONSLUSION
In this paper, a comparative study on two different wavelet transformations, namely, Haar and
Daubechies (db2) for image fusion has been presented. After completion of fusion, images were
projected into an eigenspace. Those projected fused eigenfaces are classified using a Multilayer
Perceptron. Eigenspace is constituted by the images belong to the training set of the classifier,
which is a multilayer perceptron. The efficiency of the scheme has been demonstrated on IRIS
system has achieved a maximum recognition rate of 100% using both the wavelets i.e. Haar and
Daubechies. But on an average 85% and 90.019% (using 100 testing images) and 87% and
91.5% (using 200 testing images) recognition rates were achieved for Haar wavelet and
Daubechies wavelet respectively.
Image Fusion Technique Recognition Rate
Present
method
Haar 85% When testing
image 100
87% When testing
image 200Daubechies 90.019% 91.5%
Simple Spatial Fusion[15] 91.00%
Fusion of Thermal and Visual [32] 90.00%
Abs max selection in DWT[15] 90.31%
Window based absolute maximum selection
[15]
90.31%
Fusion of Visual and LWIR + PCA [6] 87.87%

5. ACKNOWLEDGMENT
Mr. M. K. Bhowmik is thankful to the project entitled “Development of Techniques for Human
Face Based Online Authentication System Phase-I” sponsored by Department of Information
Technology under the Ministry of Communications and Information Technology, New Delhi-
110003, Government of India Vide No. 12(14)/08-ESD, Dated 27/01/2009 at the Department of
Computer Science & Engineering, Tripura University-799130, Tripura (West), India for providing
the necessary infrastructural facilities for carrying out this work. The first author would also like to
thank Prof. Barin Kumar De, H.O.D. Physics and Dean of Science of Tripura University, for his
kind support to carry out this research work.
6. REFERENCES
1. A. Pentland and T. Choudhury. “Face Recognition for Smart Environments”. IEEE Computer.
33(2):50-55, 2000
2. P. Phillips et al., The Feret Database and Evaluation Procedure for Face Recognition
Algorithms. Image and Vision Computing. May 1998. pp. 295-306
3. L. Wiskott et al., “Face Recognition by Elastic Bunch Graph Matching”. Trans. IEEE Pattern
Analysis and Machine Intelligence. 19(7):775-779. 1997
4. B. Moghaddam and A. Pentland. “Probabilistic Visual Recognition for Object Recognition”.
Trans. IEEE Pattern Analysis and Machine Intelligence. 19(7):696-710. 1997
5. P. Penev and J. Atick. “Local Feature Analysis: A General Statistical Theory for Object
Representation”. Network: Computation in Neural Systems. Mar. 1996, pp. 477-500
6. I. Pavlidis and P. Symosek. “The Imaging Issue in an Automatic Face/Disguise Detecting
System”. Proceedings of the IEEE Workshop on Computer Vision Beyond the Visible
Spectrum: Methods and Applications (CVBVS 2000)
7. Y. Adini, Y. Moses, and S. Ullman. “Face recognition: The problem of compensating for
changes in illumination direction”. IEEE Trans. On Pattern Analysis and Machine Intelligence.
Vol. 19, No. 7. pp.721-732, 1997
8. A. Ross and A. Jain. “Information fusion in biometrics”. Pattern Recognition Letters. Vol. 24,
No. 13. pp.2115-2125. 2003
9. T. K. Ho, J. J. Hull, and S. N. Srihari. “Decision combination in multiple classifier systems”.
IEEE Trans. on Pattern Analysis and Machine Intelligence. Vol. 16. No. 1. pp.66-75. 1994
10. C. Sanderson and K. K. Paliwal. “Information fusion and person verification using speech and
face information”. IDIAP Research Report 02-33. Martigny. Switzerland. 2002
11. J. Heo, S. G. Kong, B. R. Abidi and M. A. Abidi. “Fusion of Visual and Thermal Signatures
with Eyeglass Removal for Robust Face Recognition”. Conference on Computer Vision and
Pattern Recognition Workshop (CVPRW'04) Vol. 8. pp. 122-122. USA. 2004
12. I. Pavlidis, P. Symosek, B. Fritz. “A Near Infrared Fusion Scheme for Automatic Detection of
Humans”. U.S. Patent Pending. H16-25929 US. filed in September 3, 1999
13. A. Gyaourova, G. Bebis and I. Pavlidis. “Fusion of infrared and visible images for face
recognition”. Lecture Notes in Computer Science. 2004.Springer. Berlin

14. A. Selinger, D. Socolinsky. “Appearance-based facial recognition using visible and thermal
imagery: a comparative study”. Technical Report 02-01. Equinox Corporation. 2002
15. M. Hanif and U. Ali. “Optimized Visual and Thermal Image Fusion for Efficient Face
Recognition”. IEEE Conference on Information Fusion. 2006
16. G. Bebis, A. Gyaourova, S. Singh and I. Pavlidis. “Face Recognition by Fusing Thermal
Infrared and Visible Imagery”. Image and Vision Computing. Vol. 24, Issue 7. July 2006. pp.
727-742
17. A. Haar. “Zur Theorie der orthogonalen Funktionensysteme (German)”
18. http://eom.springer.de/O/o070308.htm
19. W. G. Gilbert, S. Xiaoping. “Wavelets and Other Orthogonal Systems”. 2001
20. T. Paul. “Affine coherent statesand the radial Schrodinger equation. Radial harmonic
oscillator and hydrogen atom”. Preprint
21. I. Daubechies. Ten lectures on wavelets, CBMS-NSF conference series in applied
mathematics. SIAM Ed. 1992
22. S. G. Mallat. “A theory for multiresolution signal decomposition: The wavelet representation”.
IEEE Trans. Pattern Anal. Machine Intell., vol.11. pp. 674-693. July 1989
23. Y. Meyer. Wavelets and operators. Cambridge Univ. Press. 1993
24. M. K. Bhowmik, D. Bhattacharjee, M. Nasipuri, D.K. Basu and M. Kundu. “Classification of
Polar-Thermal Eigenfaces using Multilayer Perceptron for Human Face Recognition”. in
Proceedings of the 3rd IEEE Conference on Industrial and Information Systems (ICIIS-2008).
IIT Kharagpur. India. Dec. 8-10. 2008. Page no: 118
25. D. Bhattacharjee, M. K. Bhowmik, M. Nasipuri, D. K. Basu and M. Kundu. “Classification of
Fused Face Images using Multilayer Perceptron Neural Network”. Proceedings of
International Conference on Rough sets. Fuzzy sets and Soft Computing. November 5–7,
2009. organized by Deptt. of Mathematics. Tripura University. pp. 289 – 300
26. M. K. Bhowmik, D. Bhattacharjee, M. Nasipuri, D.K. Basu and M. Kundu. “Classification of
Log Polar Visual Eigenfaces using Multilayer Perceptron”. Proceedings of The 2nd
International Conference on Soft computing (ICSC-2008). IET. Alwar. Rajasthan. India. Nov.
8–10. 2008. pp:107-123
27. M. K. Bhowmik, D. Bhattacharjee, M. Nasipuri, D. K. Basu and M. Kundu. “Human Face
Recognition using Line Features”. proceedings of National Seminar on Recent Advances on
Information Technology (RAIT-2009) Feb. 6-7,2009. Indian School of Mines University.
Dhanbad. pp: 385
28. M. K. Bhowmik. ”Artificial Neural Network as a Soft Computing Tool – A case study”. In
Proceedings of National Seminar on Fuzzy Math. & its application. Tripura University. Nov.
25 – 26, 2006. pp: 31 – 46
29. M. K. Bhowmik, D. Bhattacharjee and M. Nasipuri. “Topological Change in Artificial Neural
Network for Human Face Recognition”. Proceedings of National Seminar on Recent
Development in Mathematics and its Application. Tripura University. Nov. 14 – 15, 2008. pp:
43 – 49

30. H. K. Ekenel and B. Sankur. “Multiresolution face recognition”. Image and Vision Computing.
2005. Vol. 23. pp. 469 – 477
31. http://www.cse.ohio-state.edu/otcbvs-bench/Data/02/download.html
32. A. Singh, G. B. Gyaourva and I. Pavlidis. “Infrared and Visible Image Fusion for Face
Recognition”. Proc. SPIE. vol.5404. pp.585-596. Aug.2004
33. V. S. Petrovic and C. S. Xydeas. “Gradient-Based Multiresolution Image Fusion”. IEEE Trans.
On image Processing. Vol. 13. No. 2. Feb. 2004.
34. A. Rahim & J. V. Charangatt. “Face Hallucination using Eigen Transformation in Transform
Domain”. International Journal of Image Processing (IJIP) Vol. 3. Issue 6, pp. 265-282.
35. J. Shen and G. Strang. “Applied and Computational Harmonic Analysis”. 5(3). “Asymptotics
of Daubechies Filters, Scaling Functions and Wavelets”
36. M. K. Bhowmik, D. Bhattacharjee, M. Nasipuri, D. K. Basu and M. Kundu. “Classification of
fused images using Radial basis function Neural Network for Human Face Recognition”.
Proceedings of The World congress on Nature and Biologically Inspired Computing (NaBIC-
09). Coimbatore, India. published by IEEE Computer Society. Dec. 9-11, 2009. pp. 19-24
37. M. K. Bhowmik, D. Bhattacharjee, M. Nasipuri, D.K. Basu and M. Kundu. ”Image Pixel Fusion
for Human Face Recognition”. appeared in International Journal of Recent Trends in
Engineering [ISSN 1797-9617]. published by Academy Publishers. Finland. Vol. 2. No. 1, pp.
258–262. Nov. 2009
38. D. Bhattacharjee, M. K. Bhowmik, M. Nasipuri, D. K. Basu & M. Kundu. “A Parallel
Framework for Multilayer Perceptron for Human Face Recognition”. International Journal of
Computer Science and Security (IJCSS), Vol. 3. Issue 6, pp. 491-507.
39. Sathesh & Samuel Manoharan. “A Dual Tree Complex Wavelet Transform Construction and
its Application to Image Denoising”. International Journal of Image Processing (IJIP) Vol. 3.
Issue 6, pp. 293-300.

Fusion of Wavelet Coefficients from Visual and Thermal Face Images for Human Face Recognition – A Comparative Study

More Related Content

What's hot (19)

Similar to Fusion of Wavelet Coefficients from Visual and Thermal Face Images for Human Face Recognition – A Comparative Study (20)

Recently uploaded (20)

Fusion of Wavelet Coefficients from Visual and Thermal Face Images for Human Face Recognition – A Comparative Study