About The Organization

1 | T r a i n i n g R e p o r t ( I R I S R E C O G N I T I O N )
About the Organization (IRDE)

IRDE (Instrument Research & development Establishment):

Instruments Research & Development Establishment (IRDE), Dehradun, is an institution devoted
to research, design, development and technology transfer in the fields of sophisticated optical
and electro-optical instrumentation of vital interest to the Defence Services.
The origin of IRDE dates back to the establishment of Inspectorate of Scientific Stores in 1939 at
Rawalpindi (now in Pakistan) with responsibility to inspect telecommunication equipment used
by the Army.

VISION
To achieve excellence in the field of optics & electro-optical
Instrumentation with a commitment to provide world class
equipment
MISSION
i. To design and develop state-of-art night vision devices.
ii. To design and develop compact laser based instruments.
iii. To design and develop integrated optical and electro-
optical surveillance and fire control systems.
iv. To carry out research in the area of photonics

About the Training (Duration & Specification)

I did my training from Instrument Research & Devlopment Organization , Dehradun Uttrakhand
. It continued for about 29 days, starting from 17
th
of June 2011, till 15
th
of August 2011, under
the guidance of Mr. Amit Aran, Scientist C.

During this period I have understood the functioning and working principles of numerous
software development tools. This has developed a sense of confidence in me. And a lot of credit
goes to my guide who helped me all the way from the very beginning. My work was basically
concerned with the development of a code which could be used for the recognition of various
face images using MATLAB..

The Summer Training was divided into following modules:
1. Matlab Basics
2. Iris Recognition using JTC.
In this time period we explored the knowledge of Matlab software like Basic Matrix concepts,
plotting of functions , user interfaces, Simulink, JTC correlator, gray level images . This Training
report contains the introduction and practical implementation of the related concepts. Basically,
this Training report contains all the aspects of the things that happened during the Training
period.


(Day 1)
Photonics Division

The science of photonics includes the generation, emission, transmission, modulation ,signal
processing, switching, amplification, detection , sensing of light. The term photonics thereby
emphasizes that photons are neither particles nor waves they are different in that they have
both particle and wave nature. It covers all technical applications of light over the
whole spectrum from ultraviolet over the visible to the near-, mid- and far-infrared. Most
applications, however, are in the range of the visible and near infrared light. The term photonics
developed as an outgrowth of the first practical semiconductor light emitters invented in the early
1960s and optical fibers developed in the 1970s.

The Photonics division of IRDE is busy in achieving excellence in the field of optics & electro-
optical Instrumentation with a commitment to provide world class equipment for defense related
purposes.

The vision of the division is to;
I. To design and develop state-of-art night vision devices.
II. To design and develop compact laser based instruments.
III. To design and develop integrated optical and electro-optical surveillance and fire control
systems.
IV. To carry out research in the area of photonics.


(Day 2)
MATLAB:

MATLAB (matrix laboratory) is a numerical computing environment and fourth-generation
programming language. Developed by MathWorks, MATLAB allows matrix manipulations,
plotting of functions and data, implementation of algorithms, creation of interfaces, and
interfacing with programs written in other languages, including C, C++, Java, and Fortran.

Matlab was originally a package for matrix algebra. It has evolved to include strong graphics
abilities and an extensive programming language. It is available, in various versions, for various
types of hardware: PCs, Macintoshes, SUN workstations, Vaxs etc. On most of these systems
Matlab will be started by entering the command at the command prompt. This can however
dier, depending on the whims of your system administrator.You can interrupt, or abort,
execution of Matlab commands by entering a control C. To do this hold down the control key
and, before releasing it, press the C key.


(DAY3)
BASIC MATRIX OPERATIONS

This is a demonstration of some aspects of the MATLAB language.
First, let's create a simple vector with 9 elements called a.
a = [1 2 3 4 6 4 3 4 5]
a =

1 2 3 4 6 4 3 4 5

Now let's add 2 to each element of our vector, a, and store the result in a new vector.
Notice how MATLAB requires no special handling of vector or matrix math.
b = a + 2
b =

3 4 5 6 8 6 5 6 7

Creating graphs in MATLAB is as easy as one command. Let's plot the result of our vector
addition with grid lines.
plot(b)
grid on

MATLAB can make other graph types as well, with axis labels.
bar(b)
xlabel('Sample #')
ylabel('Pounds')

MATLAB can use symbols in plots as well. Here is an example using stars to mark the points.
MATLAB offers a variety of other symbols and line types.
plot(b,'*')
axis([0 10 0 10])

One area in which MATLAB excels is matrix computation.
Creating a matrix is as easy as making a vector, using semicolons (;) to separate the rows of a
matrix.
A = [1 2 0; 2 5 -1; 4 10 -1]
A =

1 2 0
2 5 -1
4 10 -1

We can easily find the transpose of the matrix A.
B = A'

B =

1 2 4
2 5 10
0 -1 -1

Now let's multiply these two matrices together.
Note again that MATLAB doesn't require you to deal with matrices as a collection of numbers.
MATLAB knows when you are dealing with matrices and adjusts your calculations accordingly.
C = A * B
C =

5 12 24
12 30 59
24 59 117

Instead of doing a matrix multiply, we can multiply the corresponding elements of two matrices
or vectors using the .* operator.
C = A .* B
C =

1 4 0
4 25 -10
0 -10 1


Let's find the inverse of a matrix ...
X = inv(A)
X =

5 2 -2
-2 -1 1
0 -2 1

... and then illustrate the fact that a matrix times its inverse is the identity matrix.
I = inv(A) * A
I =

1 0 0
0 1 0
0 0 1

MATLAB has functions for nearly every type of common matrix calculation.
There are functions to obtain eigenvalues ...
eig(A)
ans =

3.7321
0.2679
1.0000

... as well as the singular values.
svd(A)
ans =

12.3171
0.5149
0.1577

The "poly" function generates a vector containing the coefficients of the characteristic
polynomial.
The characteristic polynomial of a matrix A is

p = round(poly(A))
p =

1 -5 5 -1

We can easily find the roots of a polynomial using the roots function.
These are actually the eigenvalues of the original matrix.
roots(p)
ans =

3.7321

1.0000
0.2679

MATLAB has many applications beyond just matrix computation.
To convolve two vectors ...
q = conv(p,p)
q =

1 -10 35 -52 35 -10 1

... or convolve again and plot the result.
r = conv(p,q)
plot(r);
r =
1 -15 90 -278 480 -480 278 -90 15 -1

.


(Day 4)
MATLAB FUNCTIONS
Built-in functions

There are numerous built-in functions (i.e. commands) in MATLAB. We will mention a few
of
them in this section by separating them into categories.
Scalar Functions
Certain MATLAB functions are essentially used on scalars, but operate element-wise when
applied to a matrix (or vector). They are summarized in the table below.
sin trigonometric sine
cos trigonometric cosine
tan trigonometric tangent
asin trigonometric inverse sine (arcsine)
acos trigonometric inverse cosine (arccosine)
atan trigonometric inverse tangent (arctangent)
exp exponential
log natural logarithm
abs absolute value
sqrt square root
rem remainder
round round towards nearest integer
floor round towards negative infinity
ceil round towards positive infinity 14

The trigonometric functions take as input radians. Since MATLAB uses pi for the number
= 3.1415

sin(pi/2)

ans =
1

cos(pi/2)

ans =
6.1230e-017


The sine of /2 is indeed 1 but we expected the cosine of /2 to be 0. Well, remember that
MATLAB is a numerical package and the answer we got (in scientific notation) is very close
to
0 ( 6.1230e-017 = 6.123010
17
0).

Since the exp and log commands are straight forward to use, let us illustrate some of the
other commands. The rem command gives the remainder of a division. So the remainder of
12

divided by 4 is zero

rem(12,4)

ans =
0

and the remainder of 12 divided by 5 is 2.

rem(12,5)

ans =
2
The floor, ceil and round commands are illustrated below.
floor(1.4)
ans =
1

ceil(1.4)
ans =

2 15

round(1.4)

ans =
1


.(DAY 5)
MATRIX FUNCTIONS

Much of MATLABs power comes from its matrix functions. These can be further separated
into two sub-categories. The first one consists of convenient matrix building functions, some of
which are given in the table below.

eye identity matrix
zeros matrix of zeros
ones matrix of ones
diag extract diagonal of a matrix or create diagonal matrices
triu upper triangular part of a matrix
tril lower triangular part of a matrix

To create the identity matrix of size 4 (i.e. a square 4-by-4 matrix with ones on the main diagonal
and zeros everywhere else) we use the command eye.

eye(4,4)

ans =
1 0 0 0
0 1 0 0
0 0 1 0
0 0 0 1
The numbers in parenthesis indicates the size of the matrix. When creating square matrices, we
can specify only one input referring to size of the matrix. For example, we could have obtained
the above identity matrix by simply typing eye(4). The same is true for the matrix building
functions below.
Similarly, the command zeros creates a matrix of zeros and the command ones creates a
matrix of ones.

zeros(2,3)
ans =
0 0 0
0 0 0

ones(2)

ans =
1 1
1 1

We can create a randomly generated matrix using the rand command. (The entries will be
uniformly distributed between 0 and 1.)

C = rand(5,4)
C =
0.2190 0.3835 0.5297 0.4175
0.0470 0.5194 0.6711 0.6868
0.6789 0.8310 0.0077 0.5890
0.6793 0.0346 0.3834 0.9304
0.9347 0.0535 0.0668 0.8462

The commands triu and tril, extract the upper and lower part of a matrix, respectively. Let
us try them on the matrix C defined above. 19
triu(C)
ans =
0.2190 0.3835 0.5297 0.4175
0 0.5194 0.6711 0.6868
0 0 0.0077 0.5890
0 0 0 0.9304
0 0 0 0
tril(C)
ans =
0.2190 0 0 0
0.0470 0.5194 0 0
0.6789 0.8310 0.0077 0
0.6793 0.0346 0.3834 0.9304
0.9347 0.0535 0.0668 0.8462
Once the extraction took place, the empty positions in the new matrices are automatically.


(DAY6)
PROGRAMMING IN MATLAB
M-files: Scripts and functions
To take advantage of MATLABs full capabilities, we need to know how to construct long (and
sometimes complex) sequences of statements. This can be done by writing the commands in a
file and calling it from within MATLAB. Such files are called m-files because they must have
the filename extension .m. This extension is required in order for these files to be interpreted
by MATLAB.
There are two types of m-files: script files and function files. Script files contain a sequence of
usual MATLAB commands, that are executed (in order) once the script is called within
MATLAB. For example, if such a file has the name compute.m , then typing the command
compute at the MATLAB prompt will cause the statements in that file to be executed. Script
files can be very useful when entering data into a matrix.
Function files, on the other hand, play the role of user defined commands that often have input
and output. You can create your own commands for specific problems this way, which will have
the same status as other MATLAB commands. Let us give a simple example. The text below is
saved in a file called log3.m and it is used to calculate the base 3 logarithm of a positive number.
The text file can be created in a variety of ways, for example using the built-in
MATLAB editor through the command edit (that is available with MATLAB 5.0 and above), or
your favorite (external) text editor (e.g. Notepad or Wordpad in Microsoft Windows). You must
make sure that the filename has the extension .m !
function [a] = log3(x)
% [a] = log3(x) - Calculates the base 3 logarithm of x.
a = log(abs(x))./log(3);
% End of function
Using this function within MATLAB to compute log 3(5), we get
log3(5)
ans =
1.4650 28

Let us explain a few things related to the syntax of a function file. Every MATLAB function
begins with a header, which consists of the following :
(a) the word function,
(b) the output(s) in brackets, (the variable a in the above example)
(c) the equal sign,
(d) the name of the function, which must match the function filename (log3 in the above
example) and
(e) the input(s) (the variable x in the above example).
Any statement that appears after a % sign on a line is ignored by MATLAB and play ny
statement that appears after a % sign on a line is ignored by MATLAB and plays the role of
comments in the subroutine. Comments are essential when writing long functions or programs,
for clarity. In addition, the first set of comments after the header in a function serve as on-line
help. For example, see what happens when we type
help log3
[a] = log3(x) - Calculates the base 3 logarithm of x.
MATLAB gave us as help on the function we defined, the text that we included after the
header in the file.
Finally, the algorithm used to calculate the base 3 logarithm of a given number, is based on the
formula
log 3(x) = ln(|x|) / ln(3).
Since the logarithm of a negative number is undefined, we use the absolute value for safety.

LOOPS:

We will now cover some commands for creating loops, which are not only used in writing
mfiles, but in regular MATLAB sessions as well. The examples that we will give will include
both situations. The two types of loops that we will discuss are for and while loops. Both
loop structures in MATLAB start with a keyword such as for, or while and they end with the
word end.

The for loop allows us to repeat certain commands. If you want to repeat some action in a
predetermined way, you can use the for loop. The for loop will loop around some statement,
and you must tell MATLAB where to start and where to end. For example,
>> for j=1:4
j+2
end
j =
3
j =
4
j =
5
j =
6
looped through the numbers 1, , 4 and every time printed the current number plus 2.

Nested loops can also be created. In the following example, we calculate the square of the
entries in a matrix. (This again is not efficient but it is used for illustration purposes only.)
A = [1,5,-3;2,4,0;-1,6,9]
A =
1 5 -3
2 4 0
-1 6 9
for i=1:3
for j=1:3
A2(i,j) = A(i,j)^2;
end
end
A2
A2 =
1 25 9
4 16 0
1 36 81


The second type of loop is the while loop. The while loop repeats a sequence of commands
as long as some condition is met. For example, given a number n, the following m-file
(exple.m) will display the smallest non-negative integer a such that 2
a
n.
function [a] = exple(n)
% [a] = exple(n)
%
a = 0;
while 2^a < n
a = a + 1;
end
% End of function
a = exple(4)
a =
2
The conditional statement in the while loop is what makes it differ from the for loop. In the
above example we used the conditional statement
while 2^a < n
which meant that MATLAB would check to see if this condition is met, and if so proceed with
the statement that followed. Such conditional statements are also used in if statements. To
form a conditional statement we use relational operators.


(DAY7)
IMAGES & MATRICES

For any matrix X, IMAGE(X) displays a graphical image with brightness or color chosen from
the elements of X used as indices into a colormap. This demo illustrates this idea of representing
a matrix as an image and in general displaying images stored as matrices.
- The Simple Spiral Matrix
- Color map
The Simple Spiral Matrix
SPIRAL stores a simple spiral pattern into a matrix. You can see the spiral pattern of the matrix
elements in the figure. The elements of the matrix spiral away from the center, growing in
magnitude linearly. Small numbers (center values) are mapped to black and dark gray, while the
larger values (around the edge of the matrix) are mapped to light gray and white. The assignment
of small values of the matrix to black, large values of the matrix to white and intermediate values
to shades of gray determines a color map.
colormap(gray);
X = spiral(8);
image(X);


Colormaps
COLORMAP function is used to change the color mapping. The map had been set with
colormap(gray) in the previous screen. Here we change the colormap to hue-saturation-value
(hsv) color map. The colors begin with red, pass through yellow, green, cyan, blue, magenta, and
return to red.
colormap(hsv);

A completely different feature of our spiral example is revealed by the 'flag' color map. The
'flag' colormap is simply m/4 copies of the matrix flag(4), shown below, stacked on top of each
other. The colors red, white, blue and black are used cyclically as the elements of X vary and so
finer details of the image data become apparent. In this example, we can see the diagonal
patterns in the matrix
colormap(flag);

flag(4)
ans =

1 0 0
1 1 1
0 0 1
0 0 0


FUNCTIONS TO READ AND SHOW (IMAGES)

IMREAD:
Read image from graphics file
Syntax
A = imread(filename, fmt)
[X, map] = imread(...)
[...] = imread(filename)
[...] = imread(URL,...)
[...] = imread(...,Param1,Val1,Param2,Val2...)
Description
A = imread(filename, fmt) reads a grayscale or color image from the file specified by the
string filename. If the file is not in the current folder, or in a folder on the MATLAB path,
specify the full pathname.

The text string fmt specifies the format of the file by its standard file extension. For example,
specify 'gif' for Graphics Interchange Format files. To see a list of supported formats, with their
file extensions, use the imformats function. If imread cannot find a file named filename, it looks
for a file named filename.fmt.

The return value A is an array containing the image data. If the file contains a grayscale
image, A is an M-by-N array. If the file contains a truecolor image, A is an M-by-N-by-3 array.
For TIFF files containing color images that use the CMYK color space, A is an M-by-N-by-4
array. See TIFF in the Format-Specific Information section for more information.
The class of A depends on the bits-per-sample of the image data, rounded to the next byte
boundary. For example, imread returns 24-bit color data as an array ofuint8 data because the
sample size for each color component is 8 bits. See Tips for a discussion of bitdepths, and
see Format-Specific Information for more detail about supported bitdepths and sample sizes for a
particular format.
[X, map] = imread(...) reads the indexed image in filename into X and its associated colormap
into map. Colormap values in the image file are automatically rescaled into the range [0,1].
[...] = imread(filename) attempts to infer the format of the file from its content.
[...] = imread(URL,...) reads the image from an Internet URL. The URL must include the
protocol type (e.g., http://).
[...] = imread(...,Param1,Val1,Param2,Val2...) specifies parameters that control various
characteristics of the operations for specific formats.

IMSHOW:

Syntax
imshow(I)
imshow(I,[lowhigh])
imshow(RGB)
imshow(BW)
imshow(X,map)
imshow(filename)
imshow(..., param1, val1, param2, val2,...)
Description
imshow(I) displays the grayscale image I.

imshow(I,[low high]) displays the grayscale image I, specifying the display range for I in [low
high]. The value low (and any value less than low) displays as black; the value high (and any
value greater than high) displays as white. Values in between are displayed as intermediate
shades of gray, using the default number of gray levels. If you use an empty matrix ([]) for [low
high], imshow uses [min(I(:)) max(I(:))]; that is, the minimum value in I is displayed as black,
and the maximum value is displayed as white.

imshow(RGB) displays the truecolor image RGB.

imshow(BW) displays the binary image BW. imshow displays pixels with the value 0 (zero) as
black and pixels with the value 1 as white.

imshow(X,map) displays the indexed image X with the colormap map. A color map matrix may
have any number of rows, but it must have exactly 3 columns. Each row is interpreted as a color,
with the first element specifying the intensity of red light, the second green, and the third blue.
Color intensity can be specified on the interval 0.0 to 1.0.

imshow(filename) displays the image stored in the graphics file filename. The file must contain
an image that can be read by imread or dicomread. imshowcalls imread or dicomread to read the
image from the file, but does not store the image data in the MATLAB workspace. If the file
contains multiple images,imshow displays the first image in the file. The file must be in the
current directory or on the MATLAB path.
.
imshow(..., param1, val1, param2, val2,...) displays the image, specifying parameters and
corresponding values that control various aspects of the image display. The following table lists
all imshow parameters in alphabetical order. Parameter names can be abbreviated, and case does
not matter.


(DAY8)
MATLAB FFT

The fundamental tool of signal processing is the FFT, or fast Finite Fourier Transform. To take
the FFT of the sunspot data type the following.The first component of Y, Y(1), is simply the sum
of the data, and can be removed.
Y = fft(relNums);
Y(1)=[];
A graph of the distribution of the Fourier coefficients (given by Y) in the complex plane is
pretty, but difficult to interpret. We need a more useful way of examining the data in Y.
plot(Y,'ro')
title('Fourier Coefficients in the Complex Plane');
xlabel('Real Axis');
ylabel('Imaginary Axis');


The complex magnitude squared of Y is called the power, and a plot of power versus frequency
is a "periodogram".
n=length(Y);
power = abs(Y(1:floor(n/2))).^2;
nyquist = 1/2;
freq = (1:n/2)/(n/2)*nyquist;
plot(freq,power)
xlabel('cycles/year')
title('Periodogram')


(Day 11)
IRIS RECOGNITION

INTRODUCTION

This project consists of the generation of the code for the recognition of faces using various
methods of Joint Transform Correlation techniques.

Various codes were written in MATLAB for the correlation and recognition of various images.

Image Correlation

Recent technologies used for optical pattern recognition may be broadly classified into the
VanderLugt-type filter based correlation and joint transform correlation.
A VanderLugt-type correlator requires a priori fabrication of the filter used in the
correlation process, thereby prohibiting real-time operation. In addition the filter must be
accurately aligned along the optical axis in the Fourier Plane and requires close positioning
between the filter and the Fourier Transform of the input. On the other hand, a joint transform
correlator (JTC) can be operated at video frame rates and does not require the reference image to
be known substantially in advance of performing the correlation process.
One of the main problems associated with classical JTC is the presence of a strong zero-
order peak in the output plane that corresponds to the sum of autocorrelation of the reference and
the input signals and almost overshadows the desired correlation peaks.
For a single noise-free target for eg. The zero-order peak is at least four times stronger
than the crossed correlation peaks. This situation becomes more bizarre in the presence of noise
in the input scene. In a real implementation, such a zero-order peak may over saturate the output
detector and cause strong spurious reflections. In specific situations where the zero-order term is
confined to a narrow region, however, an optical stop can be effectively used at the center of the
output plane to overcome this problem.
Recently a binary JTC was proposed where the joint power spectrum (JPS) is binarized
based on a hard clipping non-linearity in the Fourier plane, only two values (+1 and -1) before
applying an inverse Fourier transform operation. When compared with the classical JTC a binary
JTC is found to yield superior correlation peak intensity correlation width and discrimination
sensitivity. The main problem with a binary JTC is the computation time required for the
determination of the threshold value used for binarizing the JPS, which acts as a constraint on
system processing speed. Also the binarization process introduces harmonic correlation peaks,

and a portion of the correlation plane energy is distributed among these higher order harmonic
terms. In addition, the higher order harmonic terms may yield false alarms or may result in
misses, thereby complicating the target detection process.
Most recently a JTC based on an amplitude modulated filter *(AMF) was reported. In
this technique, the JPS is multiplied by the AMF before the inverse Fourier transform is applied
to yield the correlation output. For single noise-free targets, the AMF based JTC is found to yield
better correlation performance than the classical and binary JTCs . However the proposed AMF
may produce high optical gain for smaller values of the reference signal power spectrum, which
may actually degrade the noise performance of the JTC. A JTC was reported that uses Fourier
plane JPS apodization.
In this technique, one uses an expensive phase-only spatial light modulator (SLM) at
the Fourier plane and the analysis results are applicable to only those reference image
JPSs that do not contain any zeros. To alleviate the problems and at the same time to
increase the auto correlation peak intensity accordingly, we propose a fringe adjusted JTC in
which a real valued filter called a fringe adjusted filter FAF is used. The performance of the
fringe adjusted JTC is investigated with computer simulation. The proposed scheme has been
found to yield better results than the classical and binary JTCs while avoiding the computation
intensive Fourier plane processing of a binary JTC.

(DAY 12)
Joint Transform Corrrelation
JTC-Brief introduction:

The joint transform correlation technique is one of the most frequently applied methods in
the field of optical classification and identification. It is used to quantify the similarity between
several input images. When realized in optics these comparisons can be effected in parallel.
Many modifications to the standard joint transform correlator (JTC) have been proposed and
verified in recent years. This project deals with an implementation of the JTC in which two types
of thresholding in the spectral domain are investigated in order to improve the performance of
the overall system.
- The JTC is an image processing technique which can be used to compare several images
in parallel. Its operation may be visualized below:

- Images for comparison are placed side by side in the input plane.
- The height of the cross-correlation peaks in the output plane represent the degree of
similarity between inputs.

Optical implementation:

- Joint transform correlation is based on two successive Fourier transforms, with some
intermediary nonlinear processing (usually a squaring operation.)
- The technique can be most efficiently effected using optical hardware.
- The Fourier transform lens (FT lens 1) will form a joint Fourier spectrum of the
input images when the input plane (a transparency) is illuminated with coherent (e.g.
laser) light.


JTC Stimulation:

- The tolerance of the JTC was investigated for rotated and dilated inputs. Two input sets
with fundamentally different image characteristics were used: fingerprints and faces.
- This example plot shows the JTC=92s sensitivity to rotated inputs for a selected element
of the fingerprint set.

- For comparison purposes the other images in the set have been correlated with the featured
image and the results plotted in the centre.


(DAY 14)
ANALYSIS OF VARIOUS TYPES OF JTCs

JOINT TRANSFORM CORRELATION:

The proposed real-time JTC used is an arrangement where the reference and the input scene are
introduced in the input plane by the use of an SLM such as a liquid crystal television. Assume
that r(x, y+y) represents the reference image and that t(x, y-y) represents the input scene in the
input plane separated by a distance 2y along the y axis. The input joint image f(x, y) can be
expressed as

f(x, y)= r(x, y+y)+ t(x, y-y) (1)
the Fourier transform of f(x, y) is computed which is given as

F(u, v)=|R(u, v)|exp[|
r
(u, v)]exp(jvy) + |T(u, v)exp[|
t
(u, v)]exp(-jvy)
(2)
Where|R(u, v)| and |T(u, v) are the amplitudes and [|
r
(u, v)] and [|
t
(u, v)] are the phases of the
Fourier transforms of r(x, y) and t(x, y), respectively, u and v are mutually independent
frequency domain variables scaled by a factor 2/, is the wavelength of the collimating light,
and f is the focal length of the Fourier transforming lenses.

The intensity of the complex light distribution produced in the back focal plane of lens1 is
called the JPS, is then detected by square law detector like a CCD array or a liquid crystal light
wave valve(LCLV), is given by

|F(u, v)|
2
=|R(u, v)|
2
+ |T(u, v)|
2
+ 2|R(u, v)| |T(u, v)| cos[|
r
(u, v) - |
t
(u, v) + 2vy]
(3)

Joint transform correlation can further be divided in three parts:-
- CLASSICAL JOINT TRANSFORM CORRELATION
- FRINGE-ADJUSTED JOINT TRANSFORM CORRELATION
- BINARY JOINT TRANSFER CORRELATION


CLASSICAL JOINT TRANSFORM CORRELATION:

In classical JTC, the JPS is Fourier transformed to yield the correlation output.

To evaluate the concept of classical JTC let us consider an example.

Here MATLAB is used to generate the base of correlation of images and to obtain their outputs.
Using one reference image and one target image of 11292 pixel we can compute the correlation
of the images and generate an output using MATLAB. These images were combined and zero
padded to form an image of 256256 pixels. The output for the two cases (same images and
different images) will be discussed ahead.

The programs generated for different faces as well as same faces are compiled in MATLAB as
shown ahead.


(DAY15)

CODE FOR SAME IMAGES:

This time the two images used are exactly the same as shown below in Figure 4(a) and (b).

Figure4 (a) Figure4 (b)

These two images are again joined in the same way as explained above to form an image as
shown in figure5.


The program for the classical joint transform correlation is shown below.

im_1=imread('C:\Documents and Settings\Guest\My Documents\face\s8\1.pgm');
figure;imshow(im_1,[]);%%%%displaying the read image
q1(1:256,1:256)=0;
q1(17:128,37:128)=im_1;
q1(129:240,129:220)=im_1;
figure;imshow(q1,[]);%%%%%%displaying the joint image

q2=fftshift(ifft2(q1));%%%%%fourier transform to obtain JPS
jps=abs(q2);
%-----------Classical JTC--------
correlation=fftshift(ifft2(jps));%%%%%fourier transform to obtain peaks
figure;mesh(abs(correlation));
view(10,10);

The output of the above program is shown in Figure 6.

Figure 6


As we can see here, the output has one zero component and two correlation peaks.
These correlation peaks show that the two images compared are similar.

CODE FOR DIFFERENT IMAGES:

Figure1 (a) Figure1 (b)

Figure1(a) is used as the reference image whereas Figure1(b) is used as the test image. The
dimensions of these images are of 11292 pixels. Both of these images are joined and zero
padded to form an image of 256256 pixels as shown in figure2

Figure 2
.

The above image is correlated according to the program shown ahead.


figure;imshow(im_1,[])%%%%displaying the read image
figure;imshow(im_2,[]);%%%%%%displaying the read image
q1(1:256,1:256)=0;
q1(17:128,37:128)=im_1;
q1(129:240,129:220)=im_2;
figure;imshow(q1,[]);%%%%%displaying the joint image
q2=fftshift(ifft2(q1)); %%%%%fourier transform to obtain JPS
jps=abs(q2);
%-----------Classical JTC--------
correlation=fftshift(ifft2(jps)); %%%%%fourier transform to obtain peaks
figure;mesh(abs(correlation));
view(10,10);

The output of the above program is obtained by a 3-D plot as shown in figure 3.

Figure 3


As we can see only one peak is obtained in the output. This peak represents the zero component.
Since there are no side peaks or Correlation peaks in the above output thus we conclude that the
input and the reference images are different.
Even though the correlation peaks are visible, another type of correlation is used instead
of Classical JTC that gives better results (sharper peaks).
This technique is known as Fringe Adjusted JTC. Which is explained ahead.

(DAY 16)

FRINGE ADJUSTED JTC (FAF JTC):

In FAF JTC the JPS is multiplied by H
faf
(u, v) before the inverse Fourier transform operation is
applied to produce the correlation output. This scheme is found to yield better results than
classical JTC. Here the fringe adjusted filter (FAF) is defined as

H
faf
(u, v) = _____B(u, v)____ (5)
A(u, v) + |R(u, v)|
2

Where A(u, v) and B(u, v) are either constants or functions. When B(u, v) is properly selected,
one can avoid having an optical gain greater than unity. With a very small value of A(u, v), the
pole problem is overcome, while at the same time it is possible to achieve a very high
autocorrelation peak. The function A(u, v) , may be used to suppress noise or band limit the
signal or both. For example, if the noise power spectrum is known, the A(u, v) factor may be
chosen to suppress the noise spectrum at the Fourier plan. Therefore, in a fringe adjusted JTC,
the amplitude matching is used more effectively to produce sharper and larger correlation peak
intensity.

Notice that the filters used in VanderLugt type correlators involve both magnitude and phase,
thus complicating the filter fabrication process. On the other hand, the FAF is a real valued
function because it involves only the intensity (i.e., the JPS) and has no phase terms. Therefore a

FAF is more suitable for optical implementation. Also, the computations involving the FAF may
be completed long before the input scene is introduced in the input plane of the JTC. Thus the
inclusion of the filter does not have any significant detrimental effect on the processing speed of
the system. However, an additional spatial light modulator is necessary to display the FAF
function.

The fringe adjusted JPS is obtained by multiplying the filter function with the JPS. This
multiplication is achieved by displaying the JPS and the FAF in two separate SLMs placed side
by side and then illuminating the SLMs with the same laser, using a beam splitter and mirror
combination. Thus the Fringe adjusted filter may be expressed as

G(u, v) = H
faf
(u, v) |F(u, v)|
2

[B(u, v)/{A(u, v) + |R(u, v)|
2
] {|R(u, v)|
2
+ |T(u, v)|
2
+ 2|R(u, v)| |T(u, v)| cos[|
r
(u, v)
- |
t
(u, v) + 2vy]} (6)

When B(u, v) = 1 and |R(u, v)|2 >> A(u, v), the FAF approaches a perfect real-valued inverse
filter, and eq.(5) is given by

G(u, v) ~ 2{1 + cos[|
r
(u, v) - |
t
(u, v) + 2vy]} (7)
The program for correlation same as well as different images using FAF JTC is discussed ahead.

CODE FOR SAME IMAGES:

figure;imshow(im_1,[]);%%%%%displaying the read image
figure;imshow(im_1,[]);
q1(1:256,1:256)=0;
q11(1:256,1:256)=0;
q1(17:128,37:128)=im_1;
q1(129:240,129:220)=im_1;
q11(17:128,37:128)=im_1;
figure;imshow(q1,[]);

for ic=1:256;%%%%%multiplying with the FAF function
for jc=1:256
q23(ic,jc)=1/(.000001+abs(q22(ic,jc)));
end
end
jps=abs(q2);
jps22=jps.*q23;
%----------FAF JTC---------
correlation22=fftshift(ifft2(jps22)); %%%%%fourier transform to obtain peaks
figure;mesh(abs(correlation22));
view(10,10);

Figure5(a) Figure5(b)

The above images are exactly same and of the same size (11292 pixels).
These are combined to form an image of 256256 pixels as shown in figure 6


Figure 6.

The corresponding output is obtained on a 3-D graph as shown in figure 7.

Figure 7.

From the output graph we can comprehend that the two images are same, due to the presence of
the correlation peaks. As we can see, the peaks obtained in FAF JTC are sharper and larger than
those obtained in Classical JTC.

CODE FOR DIFFERENT IMAGES:

q1(1:256,1:256)=0;
q11(1:256,1:256)=0;

q1(17:128,37:128)=im_1;
q1(129:240,129:220)=im_2;
q11(17:128,37:128)=im_1;
q2=fftshift(ifft2(q1));
for ic=1:256;
for jc=1:256
q23(ic,jc)=1/(.000001+abs(q22(ic,jc)));
end
end
jps=abs(q2);
jps22=jps.*q23;
%----------FAF JTC---------
correlation22=fftshift(ifft2(jps22));
view(10,10);

The input and the reference images used here are shown in figure 8(a) and (b)

Figure8(a) Figure8(b)


The images are combined similarly as done above to form an image of 256256 pixels as shown
in figure 9.

Figure 9

The output of the correlation of the above images is shown in figure 10.
Figure 10.


As we can see from the output that the correlation peaks are absent. This shows that the images
are different.


(DAY 18)
BINARY JTC

In a binary JTC the JPS is first binarized according to a threshold value before taking the
inverse Fourier transform of the JPS.
+1 if ,F (u, v),
2
> T
f

,F (u, v),
2
=
0 otherwise. (9)

Where T
f
is the JPS binarization threshold, defined by

T
f
= mean ,F (u, v),
2
(10)

CODE FOR BINARY JTC:

clear all
clc
im_1=imread('C:\Documents and Settings\Guest\My
Documents\face\s8\1.pgm');
im_1=imread('C:\Documents and Settings\Guest\My
Documents\face\s8\1.pgm');
q1(1:256,1:256)=0;
q11(1:256,1:256)=0;
q1(17:128,37:128)=im_1;
q1(129:240,129:220)=im_1;
q11(17:128,37:128)=im_1;
for ic=1:256;
for jc=1:256
q23(ic,jc)=1/(.000001+abs(q22(ic,jc)));
end

end
jps=abs(q2);
jps22=jps.*q23;
%----------FAF JTC---------
correlation22=fftshift(ifft2(jps22));
view(0,0);

%%%%-----BINARIZATION PROCESS------

bjps(1:256,1:256)=0;
s=sum(sum(jps22));
avg=s/(256*256);
for ic=1:256;
for jc=1:256
if jps22(ic,jc)>avg
bjps(ic,jc)=1;
else
bjps(ic,jc)=0;
end
end
end
correlation23=fftshift(ifft2(bjps));% DISPLAY THE RESULTS
view(0,0);

Figure 11 Figure 12

Figure 11 and figure 12 show the input ant reference image respectively.


Figure 13

Figure 14

As it can be observed that the two correlation peaks are here intensified with respect to the
average component. This is done by taking the mean of the pixels and binarizing them with
respect to the mean value of the pixels as explained by equation (9) and (10).


(DAY14)

BINARY DIFFERENTIATED JTC -BDJTC:

Optical information-processing techniques for pattern recognition have generated
considerable interest in the optics community in the last few decades. These are two widely used
architectures to implement correlation-based optical pattern recognition systems, known as the
VanderLugt correlator (VLC) and the joint transform correlator
(JTC). The VLC requires a priori fabrication of the filter, while for the JTC one does not need to
know the reference image beforehand. It is possible to have real-time operation at video frame
rates with a JTC.

Although the list of advantages associated with a JTC configuration is very long, there are
some problems also attached to it. One of its major drawbacks is the presence of a broad zero-
order (dc) peak in the output plane. There have been many attempts to overcome this
shortcoming. Implementation of an adaptive real-time JTC architecture using liquid-crystal TVs
was done. In such a system, optical or electronic techniques could be employed to subtract the dc
pattern from the correlation plane. Demonstration of an incoherent image subtraction technique
using electron-trapping materials was done. An electron-trapping film could be coupled directly
with a CCD chip to perform parallel real-time image subtraction between two simultaneous
scenes or subsequent frames. A widely used technique for dc removal is the subtraction of the
power spectrum of the input scene only and the power spectrum of the reference image only
from the joint power spectrum (JPS).

Later a binary JTC (BJTC)was presented, in which the JPS is binarized by using nonlinearity
at the Fourier plane before applying the final inverse Fourier transformation. In comparison with
the classical JTC (CJTC) a BJTC produces a higher correlation intensity, higher discrimination
ability, and narrower correlation width. The main bottleneck of a BJTC is the computation of the
threshold value used to binarize the JPS, hindering it from real-time operation.

The use of differential (or gradient) correlation is much simpler alternative idea which has
been introduced to attain high discrimination. A preprocessed JTC was introduced whose input
image is preprocessed by differential-like operators. Later a differential JTC (BJTC) was
proposed in which finite difference operation is applied to the JPS. The differencing, being a
high-pass operation, eliminates the unwanted autocorrelation dc spot formed at the center of the
correlation peaks, and thus enhances the peaks. They also showed that the binary version of the
DJTC offers higher discrimination ability and lower computation cost than the BJTC presented
by earlier proposals. Differential and binary differential JTCs are found to be associated with
high tolerance for illumination variation of the reference and target. This makes them suitable for
illumination-independent pattern recognition applications.

Most of the existing configurations of the JTC use two input images- one target and one
reference. They utilize only 50% of the space-bandwidth product in the correlation plane. The

use of multiple input images may enable additional functions to be performed by using a larger
space-bandwidth product. JTCs using multiple input images encoded in the spatial domain has
been reported by many researchers. In all these studies, removal of dc remains a major problem,
and again various techniques have been applied for this purpose.
In this section a binary differential JTC is demonstrated. In real-time situations the input scene
is captured with a CCD or thermal camera. Thermal cameras are widely used for night vision
imaging; hence images of reduced contrast considerably degraded by background noise are
obtained. The joint input images are optically Fourier-transformed to obtain JPS. The JPS is the
first differentiated and then binarized before inverse Fourier
Transformation. Differential processing of the JPS removes the dc and hence improves the
detection efficiency. There is also a study done on the effect of Gaussian noise on the correlation
output in both the CJTC and BDJTC and with single and multiple targets.

(DAY 19)

Binary Differential Joint Transform Correlator:

Let f(x, y) be the input images to a JTC, consisting of the reference image r(x-a, y) and the
target image r(x+a, y) separated by a distance 2a.

f(x, y) = r(x+a, y) + r(x-a, y). (1)

The Fourier transform of f(x, y) is captured by an intensity-sensing device, a CCD camera:
I (u, v) =|F (u, v) |
2
= T
2
(u, v) + R
2
(u, v) + R(u, v)T
*
(u, v)exp(-j2au) +
R
*
(u, v)T(u, v)exp(j2au) (2)

Where (u, v) are the frequency-domain coordinates and F(u, v), T(u, v) and R(u, v) are the
Fourier transforms of the f(x, y), t(x, y) and r(x, y) respectively. The asterisk denotes the
complex conjugate. Taking one more Fourier transform of eq. (2) gives the output of the JTC.
C(x, y) = t(x, y) t(x, y) + r(x, y) r(x, y) + r(x-2a, y) t(x-2a, y) + r(x+2a, y) r(x+2a,
y) (3)

Where the symbol denotes the correlation operator.


The first two terms of eq. (3) give rise to dc, whereas the thirds and fourth terms are the
correlation terms. The first and second in eq. (2) the main sources of dc, are the slowly changing
parts of the JPS. The third and fourth terms, being the interference terms, mainly carry the
correlation information. Thus by removing the dc, i.e., the first and second terms of eq(2) the
detection efficiency of the system will be improved. Various techniques for the dc removal have
been reported ahead, such as dc blocking, binarizing the JPS, and subtraction.

Although the differential operation is difficult to realize optically, it can be implemented
digitally with less computation than in the case of a BJTC based on adaptive, median, or logical
median thresholding. The DJTC will have a bipolar JPS. A binarized version of the DJTC can be
easily generated by observing the sign changes in the JPS. Being lighter efficient, it is found to
yield higher correlation peak intensity. Since the differential operation can depress slow
variations, it is introduced to the JPS to improve the contrast. In discretely sampled systems, a
partial differential can be approximated by first order forward difference. Since the derivative
always assumes a maximum in the direction of an edge (i.e., the direction perpendicular to the
fringes), the derivative along the u direction (line joining the reference and target objects) has
been taken. A partially differentiated JPS along the u axis is given by

t
-1
{oI(u ,v)/ou } = (-jx) t
-1
{ I(u, v) }, (4)

Where x id the coordinate of the output plane t
-1
denoted the inverse Fourier transformation,
and the differential property of the Fourier transform has been used. The output of the JTC after
square law is given as

|(-jx) t
-1
{I(u,v)}|
2
= x
2
| C(x,y)|
2
(5)

Where C(x, y) given in eq(3), is the output of the conventional JTC. From eq(5) we infer that
differential operation of the JPS is equivalent to the operation of putting a mask on the output
plane whose transmittance is proportional to x
2

For binary JTCs, a threshold value is needed to binarize the JPS of the reference signal and the
input scene. The subset median threshold method has been used for the threshold binarization of
the differential JPS. The threshold value has to be updated for every new input scene. In subset
median thresholding, the JPS is segmented spatially in the Fourier plane. The median of each
subset is computed, and then each subset is binarized according to its subsets of n X m pixels,
where n N and m M. the threshold value for each subset is defined as the median of that
segment:

T
nm
= median [C
nm
(u, v)] (6)


Where C
nm
(u, v) is the segmented JPS. Compared with median thresholding, the subset median
threshold method is computationally less intensive; the complexity of median computation is
reduced as the segmentation size is decreased. A binary version of a DJTC, viz. BDJTC which is
more efficient, can be obtained after binarizing the bipolar differential JPS retaining its sign, as
follows,

I (u, v) = ( +1 ,C
u
(u, v)> T
nm
, (7)
-1 ,C
u
(u, v) T
nm
,

A BDJTC inherits all the characteristics of a DJTC, but provides sharper and more intense
correlation, but provides sharper and more intense correlation peaks.

(DAY22)

CODE FOR BDJTC(Single object):

A program in MATLAB is written for implementing the Single object binary differential joint
transform correlator and the following results are obtained, notice the peak intensity that is
being sharpened by the binarization and the average component disappears due to differentiation
process.

im_1=imread('C:\matlab7\work\face\s8\1.pgm');
figure;imshow(im_1,[]);%%%%%%%%%%%% displaying the read image
figure;imshow(im_2,[]);%%%%%% displaying the read image
q1(1:256,1:256)=0;
q1(17:128,37:128)=im_1;
q1(129:240,129:220)=im_2;
figure;imshow(q1,[]);%%%%%%%%%%%%%%% displaying the joint image
q2=fftshift(fft2(q1));%%%%%%%%%%%%%%%5taking Fourier transform%
jps=abs(q2);
djps=diff(jps);%%%%%%%%%%%%%STEP 1:DIFFRENTIATE

med=median(median(djps));A
bdjps(1:256,1:256)=0;%%%%%%%%%%%%%%%%STEP 2:BINARIZE
for ic=1:112
for jc=1:92
if djps(ic, jc)> med
bdjps(ic, jc)=1;
else
bdjps(ic, jc)=-1;
end
end
end
correlation3=fftshift(ifft2(bdjps));
view(10,10);

Figure 15 and 16 represent the reference image and the joint image respectively.

Figure 15:Single image Figure 16: Joint image


Figure 17: Single Object BDJTC
From the above image we can see that the Zero component(dc value) is reduced to zero and the
correlation peaks are obtained which are higher and better in intensity than the peaks obtained in
the Classical JTC.


(DAY23)
CODE FOR BDJTC(Multiple objects):

A program in MATLAB is written for implementing the Multiple object binary
differential joint transform correlator and the following results are obtained, notice the peak
intensity that is being sharpened by the binarization and the average component disappears due
to differentiation process. Notice the six peaks obtained as two object images are taken.

q1(1:256,1:256)=0;
q1(17:128,37:128)=im_1;
q1(76:187,129:220)=im_2;
q1(129:240,37:128)=im_1;
figure;imshow(q1,[]);%%%%%%%%%%%%%%% displaying the joint
multiple image
q2=fftshift(fft2(q1));%%%%%%%%%%%%%%%taking fourier transform%
jps=abs(q2);
djps=diff(jps);%%%%%%%%%%%%%STEP 1:DIFFRENTIATE
med=median(median(djps));
bdjps(1:256,1:256)=0;%%%%%%%%%%%%%%%%STEP 2:BINARIZE
for ic=1:112
for jc=1:92
if djps(ic, jc)> med
bdjps(ic, jc)=1;
else
bdjps(ic, jc)=-1;
end
end
end
correlation3=fftshift(ifft2(bdjps));%%%%%%%%%obtaining peaks
view(10,10);

Figure 18 shows the image that is to be recognized and Figure 19 shows the correlation of the
multiple input images that are joined and zero padded to form an image of 256256 pixels.


Figure 18: Single image Figure 19: Joint image

Figure 20:3-D Output

From the output graph we can see the various correlation peaks that are obtained.

(DAY25)
CONCLUSION

In this project we have implemented the CJTC,FAF JTC ,BJTC and BDJTC with input synthetic
images. The idea of differential processing and binarization as proposed was employed. The
subset median threshold method was used as the threshold value for binarization of the
differential JPS. Differential processing of the JPS removes the zero-order spectra and hence
improves the detection efficiency. The differential and binary differential operations are
performed digitally. It is found that BDJTC performs better than the CJTC. Also, binary and
binary differential JTCs are found to be associated with high tolerance for illumination variation
of the reference and target. Experiments with different reference objects and single and multiple
target objects have been presented.

We have presented a fringe-adjusted filter based JTC for target detection. This technique is
found to yield substantially better correlation output than the classical and binary JTCs for input
scenes involving single as well as multiple objects. The FAF is designed such that it avoids the
problems associated with and inverse filter, while producing a high autocorrelation peak
intensity. It may also be used to attenuate the noise that is present in the input scene provided
that the factor A(u, v) is selected properly. Computer simulation results show that the Fringe
adjusted JTC yields better correlation peak intensity. For input scenes involving multiple objects
where two or more objects are identical, false autocorrelation peaks may be produced at the
output plane. This problem may be alleviated by subtracting the input- only JPS from the JPS at
the expense of an additional processing step. The input only JPS may be obtained by displaying
only input scene in the input plane SLM in the absence of reference image then recording the
JPS. By using the proposed technique, we are able to avoid the computation intensive JPS
binarization process associated with binary JTC and the false alarms and misses often associated
with the multi object binary JTC because of JPS binarization. The fringe adjusted JTC, however,
requires an extra SLM to display the FAF filter. Liquid crystal televisions may be used for this
purpose. Note that the computation associated with the FAF can be completed long before the
input scene is introduced into the input plane. Therefore the use of this additional filter may not
have any detrimental effect on the system-processing speed.



Motivating Factors

The reason for my selecting IMAGE RECOGNITION for the Summer Training is my keen
interest in BIOMETRICS. It all started in the 6
th
semester when we had to study a subject named
INFORMATION SECURITY.

When I studied that subject thoroughly during the preparation of the semester exams, it all
started fascinating me.HOW we can authenticate a user to access the data without any security
breaches and how iris is used as a biometrics etc etc . As I went deeper and deeper, I came to
know about various biometrics, which was the turning point. By then Is sure that I need to
utilize my summer vacations and get deeper into this fascinating world of Biometrics.

Also I had a chance to discuss with some of my friends faculty and family about my Summer
Training, and I coincidently went through a conversation with one of my senior who actually did
his Summer Trainingfrom DRDO. After discussing with him about the details of the course & its
benefits, I became sure about my Summer Training at DRDO and eventually I succeeded in
doing so.

Since IRDE,DRDO is under Ministry Of Defense so its a great opportunity to work with such
an organisation .


Whats new?

It was really a totally different experience doing my training there at the IRDE DRDO ,
dehradun. The environment was completely different. We also studied the things here at the
college, , but there, it was a bit different experience.

Here at college, we studied about the basic concepts of biometrics , bt there I actually come to
know how iris is used as biometric .The thing at college is we r only doing the theory thing but in
ma training period at IRDE i come to know how things go practically. How a software ie
MAtlab works and how using that software we can write coding for the iris scanning, various
computational works are the best part of the training.

Also the level of the training was very good. It all went in a very efficient and lucid manner, so
that each and everybody were able to understand and grasp the things very easily. The level of
the study material and the presentation was also up to the mark. They provided a good
environment for enhancing our skills.


Journey & Environment

I got a chance to do my training from IRDE DRDO, DEHRADUN, just because of the presence
of myuncle there. He knows some scientist at IRDE bcause of that only I got the opportunity to
do training there , and I am very thankful to him.Because of this, I got a chance to be there this
summer, and also join a good organization for my training.

When I went there, the environment I found was totally different from the environment we find
here in our college. It was a combination of that of an educational institution as well as the
corporate environment. We were provided much of the practical knowledge than the theory part.

In this training period, I also learnt how to manage ourselves in a corporate environment. I came
to know about the quality a professional should have.


Conclusion

Its a really great experience to do my trainingat IRDE. Its also very useful to do the training
under the guidance of a person having a good experience as well as good knowledge. The
training with the HCL should also be very useful in the perspective of both knowledge, and the
career. In the case of career, it would be an important one because it has the tag of a organization
like IRDE, and in the case of knowledge, it is important because in this training I studied the
concepts of Photonics(IMAGE PROCESSING) , which is the most important part of any
organization.

About The Organization

Enviado por

Dados do documento

Descrição original:

Título original

Direitos autorais

Formatos disponíveis

Compartilhar este documento

Compartilhar ou incorporar documento

Opções de compartilhamento

Você considera este documento útil?

Este conteúdo é inapropriado?

Direitos autorais:

Formatos disponíveis

About The Organization

Enviado por

Direitos autorais:

Formatos disponíveis

1 | T r a i n i n g R e p o r t ( I R I S R E C O G N I T I O N )

About the Organization (IRDE)

Você também pode gostar