IRDE is an institution devoted to research, design, development and technology transfer in the fields of sophisticated optical and electro-optical instrumentation. 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.
IRDE is an institution devoted to research, design, development and technology transfer in the fields of sophisticated optical and electro-optical instrumentation. 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.
Direitos autorais:
Attribution Non-Commercial (BY-NC)
Formatos disponíveis
Baixe no formato DOCX, PDF, TXT ou leia online no Scribd
IRDE is an institution devoted to research, design, development and technology transfer in the fields of sophisticated optical and electro-optical instrumentation. 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.
Direitos autorais:
Attribution Non-Commercial (BY-NC)
Formatos disponíveis
Baixe no formato DOCX, PDF, TXT ou leia online no Scribd
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 2 | 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 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.
3 | 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 )
(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.
4 | 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 )
(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.
5 | 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 )
(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 6 | 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 )
MATLAB can make other graph types as well, with axis labels. bar(b) xlabel('Sample #') ylabel('Pounds') 7 | 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 )
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]) 8 | 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 )
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' 9 | 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 )
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
10 | 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 )
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 11 | 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 )
... 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 12 | 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 )
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 13 | 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 )
.
14 | 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 )
(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
15 | 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 )
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
16 | 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 )
.(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) 17 | 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 )
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.)
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.
18 | 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 )
(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 19 | 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 )
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. 20 | 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 )
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
21 | 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 )
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.
22 | 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 )
(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);
23 | 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 )
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); 24 | 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 )
flag(4) ans =
1 0 0 1 1 1 0 0 1 0 0 0
25 | 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 )
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: 26 | 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 )
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.
27 | 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 )
(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');
28 | 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 )
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')
29 | 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 )
(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, 30 | 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 )
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.
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: 31 | 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 )
- 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.
32 | 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 )
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.
33 | 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 )
(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
Joint transform correlation can further be divided in three parts:- - CLASSICAL JOINT TRANSFORM CORRELATION - FRINGE-ADJUSTED JOINT TRANSFORM CORRELATION - BINARY JOINT TRANSFER CORRELATION
34 | 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 )
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.
35 | 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 )
(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.
36 | 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 )
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
37 | 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 )
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.
38 | 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 )
im_1=imread('C:\Documents and Settings\Guest\My Documents\face\s8\1.pgm'); figure;imshow(im_1,[])%%%%displaying the read image im_2=imread('C:\Documents and Settings\Guest\My Documents\face\s1\1.pgm'); 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
39 | 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 )
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
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 40 | 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 )
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
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:
im_1=imread('C:\Documents and Settings\Guest\My Documents\face\s8\1.pgm'); figure;imshow(im_1,[]);%%%%%displaying the read image im_1=imread('C:\Documents and Settings\Guest\My Documents\face\s8\1.pgm'); 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,[]); figure;imshow(q11,[]); q2=fftshift(ifft2(q1)); %%%%%fourier transform to obtain JPS q22=fftshift(ifft2(q11)); %%%%%fourier transform to obtain JPS 41 | 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 )
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
42 | 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 )
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:
im_1=imread('C:\Documents and Settings\Guest\My Documents\face\s8\1.pgm'); figure;imshow(im_1,[]); im_2=imread('C:\Documents and Settings\Guest\My Documents\face\s1\1.pgm'); figure;imshow(im_2,[]); q1(1:256,1:256)=0; q11(1:256,1:256)=0; 43 | 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 )
q1(17:128,37:128)=im_1; q1(129:240,129:220)=im_2; q11(17:128,37:128)=im_1; figure;imshow(q1,[]); figure;imshow(q11,[]); q2=fftshift(ifft2(q1)); q22=fftshift(ifft2(q11)); 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)); figure;mesh(abs(correlation22)); view(10,10);
The input and the reference images used here are shown in figure 8(a) and (b)
Figure8(a) Figure8(b)
44 | 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 )
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.
45 | 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 )
As we can see from the output that the correlation peaks are absent. This shows that the images are different.
46 | 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 )
(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'); figure;imshow(im_1,[]); im_1=imread('C:\Documents and Settings\Guest\My Documents\face\s8\1.pgm'); 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,[]); figure;imshow(q11,[]); q2=fftshift(ifft2(q1)); q22=fftshift(ifft2(q11)); for ic=1:256; for jc=1:256 q23(ic,jc)=1/(.000001+abs(q22(ic,jc))); end 47 | 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 )
end jps=abs(q2); jps22=jps.*q23; %----------FAF JTC--------- correlation22=fftshift(ifft2(jps22)); figure;mesh(abs(correlation22)); 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 figure;mesh(abs(correlation23)); view(0,0);
Figure 11 Figure 12
Figure 11 and figure 12 show the input ant reference image respectively.
48 | 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 )
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).
49 | 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 )
(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 50 | 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 )
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.
51 | 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 )
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)
52 | 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 )
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 im_2=imread('C:\matlab7\work\face\s8\1.pgm'); 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 53 | 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 )
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)); figure;mesh(abs(correlation3)); view(10,10);
Figure 15 and 16 represent the reference image and the joint image respectively.
Figure 15:Single image Figure 16: Joint image
54 | 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 )
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.
55 | 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 )
(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.
im_1=imread('C:\matlab7\work\face\s8\1.pgm'); figure;imshow(im_1,[]);%%%%%%%%%%%% displaying the read image im_2=imread('C:\matlab7\work\face\s8\1.pgm'); figure;imshow(im_2,[]);%%%%%%%%%%%% displaying the read image 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 figure;mesh(abs(correlation3)); 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.
56 | 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 )
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. 57 | 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 )
(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.
58 | 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 )
59 | 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 )
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 .
60 | 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 )
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.
61 | 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 )
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.
62 | 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 )
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.