3.   FACEFINDER
3.1   Introduction
3.1.1   Expectation
3.1.2   FaceFinder Input
3.1.3   FaceFinder
3.1.4   Restrictions
3.2   Identification
3.2.1   Operation
3.2.2   Template Matching Output
3.2.3   Checking Identification Results
3.2.4   Analysis
3.2.5   Examples
3.2.6   Identify with the Template File
3.3   Verification
3.3.1   Operation
3.3.2   Example

3.   FaceFinder

This is an example of a customized version of the ImageFinder, applied to the Feret Facial Image Database. All parameters are eliminated.

Figure 3.1   Example Menu.

To start the FaceFinder in the ImageFinder, click menu �Example/FaceFinder�.
 

Figure 3.2   The Feret Database.

3.1   Introduction

3.1.1   Expectation

With image matching, the face images are converted into templates first. This one-time conversion runs at the speed of several images per second. After that, the template matching runs at a speed of higher than 100,000 matches per second and it will eliminate about 98% - 99% of unmatched images. This means if you want to match one newly captured image against 1,000,000 existing images, in a matter of seconds, the FaceFinder will reduce the 1,000,000 possible matches to about 10,000 candidates. At this point, the other types of matching can further narrow down the retrieved set.
 

Figure 3.3   The FaceFinder.

There is only one example in the FaceFinder, with following results:

Total Images N = 1,032 = 516 pairs
Possible Matches N*N = 1,065,024
Attrasoft Matches = 14,300

Actual Duplicates = 2,064
Attrasoft Found Duplicates = 2,064

Percent Eliminated = 98.7 %
= 1 � 14,300/1,065,024
Percent of Duplicates Found = 100 %
= 2,064/2,064

This example has 516 pairs of face images or 1,032 images. In the N x N Match, each image matches itself and its partner in the pair, giving a total of 2 * N = 2 * 1,032 = 2,064 positive matches.

3.1.2   FaceFinder Input

There is only one way to enter data into the FaceFinder, by input file. The FaceFinder will compare each of the images in the file to all other images in the file. Clicking the �File Input� button and selecting a file can specify the input file.

 The input files list one image per line. Each line specifies an absolute path. For example,

C:\xyz1\0001.jpg
C:\xyz1\0002.jpg
C:\xyz2\0003.jpg
C:\xyz2\0004.jpg
�
There is one file for Identification in the software CD, �facefinder_input_id.txt�, and there is one file for Verification, �facefinder_b1_first10.txt�.

3.1.3   FaceFinder

The FaceFinder is an example of a customized version of the ImageFinder. There are many parameters in the ImageFinder. The FaceFinder has eliminated all of the parameters.

The image matching will be done in three steps:

Let us look at each phase.

Training/Initialization

Initialization loads the ImageFinder parameters. Training uses the data collected in advance. Both steps are combined together and done in a single click.
Converting Images to Records:
This step is slow (several images per second); however:

(1) This step can be done once for all; and
(2) This is linear, i.e. the time is directly proportional to the number of images.

Therefore, this step does not have much impact on the operating speed. If you use the default images, the records are loaded when you click the �Initialize� button.

Template Matching
The matching speed will be between 100,000 and 1,000,000 comparisons per second.


3.1.4   Restrictions

The restrictions are:

Feret database;
Front view;
Distance between the Camera and the face is fixed;
Light condition is fixed;
Image dimension is 256 x 384;
Image type is jpg.


3.2   Identification

The image Identification will be done in three steps:


3.2.1   Operation

The Identification procedure is:

Now let us look at the details.

1. Initialization and Training

Initialize and train the software by clicking �Initialize� button. This button will first initialize the software, which simply means to load the parameters; then train the software, which is similar to the ImageFinder. At the end of this step, you should see:
 
Parameters loaded!
bioFilter loaded!
�
Neural Filter loaded!
bioFilter 2 loaded!
Neural Filter 2 loaded!
2. Entering Data
Select input file by clicking the �File Input� button. In this example, the input file for the Identification example is �.\facefinder_input_id.txt�. Select this file. This file has 1,032 images and the software will go through each one of them to make sure the files exist. This will take a while and at the end of this process, you should see:
Number of blocks = 1
Block 0. Length = 1032
C:\�\00019ba010_960521.jpg
C:\�\00019bj010_960521.jpg
C:\�\00029ba010_960521.jpg
C:\�\00029bj010_960521.jpg
�
Buttons:
  •  To see the original content of this file, click the �ShowFile� button.
  •  To clear the text window, click the �Clear� button.
  •  To see the first image in the input file, click the �First� button.
  •  To see the next image in the file, click the �Next� button. If you keep clicking the �Next� button, the FaceFinder will scan through all images.
    • 3. Templates
    If you use the default images, this step is optional because the records are loaded when you click the �Initialize� button.

    Convert images in the input file to records by clicking the �Templates� button. The results are saved to a1.txt. Once the image is converted into internal templates, you do not have to convert them again. This step is slow (several images per second); however:

  •  This step can be done once for all; and
  •  This is linear, i.e. the time is directly proportional to the number of images.
    • Therefore, this step does not have much impact on the operating speed.


    4. Template Matching

    Identify matches by clicking the �Identify� button. The FaceFinder will compare each of the images in the file to all other images in the file. If there are N images in the input file, there will be N*N comparisons.


    5. Check the Results

    Check the results of the template matching by clicking the �Results� button. The checking result  will be printed to the text window; and it will indicate that the number of matches found in b1.txt will agree with the specifications in the b1_matchlist.txt. This will quickly let you know the positive identification rate. The matching information is stored in file, �fdacefinder_b1_matchlist.txt�.


    3.2.2   Template Matching Output

    The template matching runs at a speed of higher than 100,000 matches per second and it will eliminate about 98% - 99% of the unmatched images.

    This means if you want to match one newly captured image against 1,000,000 existing images in a matter of seconds, the FaceFinder will reduce the 1,000,000 possible matches to about 10,000 candidates.

    The result of matching each of the 1,032 images against the 1,032 images is in the file b1.txt. At the end of the matching, b1.txt will be opened for you. The b1.txt will look like this:

    C:\...\data1\00019ba010_960521.jpg
    C:\...\data1\00019ba010_960521.jpg
    C:\...\data1\00019bj010_960521.jpg

    C:\...\data1\00019bj010_960521.jpg
    C:\...\data1\00019ba010_960521.jpg
    C:\...\data1\00019bj010_960521.jpg

    �

    C:\...\data1\01209fb010_940128.jpg
    C:\...\data1\01063ba010_960521.jpg
    C:\...\data1\01063bj010_960521.jpg
    C:\...\data1\01209fa010_940128.jpg
    C:\...\data1\01209fb010_940128.jpg

    Total Number of Matches = 14,300

    This file is divided into blocks separated by a blank line. Each image has its own block. A block looks like this:
    C:\...\data1\00019ba010_960521.jpg
    C:\...\data1\00019ba010_960521.jpg
    C:\...\data1\00019bj010_960521.jpg


    The first line in this block is the input for Identification, the rest of the lines are the output for Identification. In the above example, 00019ba010_960521.jpg matches with itself and 00019bj010_960521.jpg.

    3.2.3   Checking Identification Results

    If this is a test run, i.e. you know the correct answers; you can see the matching results in seconds. You must prepare a file, which indicates the matching pairs. To test the results in b1.txt, you must prepare B1_matchlist.txt file.

    An example of b1_matchlist.txt is:

    5
    1 IMAGE00053770 IMAGE01312024
    2 IMAGE00053771 IMAGE01312025
    3 IMAGE00053772 IMAGE01312026
    4 IMAGE00053773 IMAGE01312027
    5 IMAGE00053774 IMAGE01312028

    Line 1 is the �Line number� in this file. The �Line Number� does not have to be in order. Each line has the following format:

    Number, tab, filename, tab, filename, tab.

    Note:
    (1) You must have a tab at the end of each line;
    (2) The file names do not contain �.jpg�.

    Once you get the two files prepared, click the �Results� button.

    There are two common errors:

    (1) The last Tab is missing;
    (2) The number of rows is less than the first number in the file.

    Check the results of the template matching by clicking the �Results� button. You will see the something like the following in the text window:

    Checking Template Matching Results!
    Get b1.txt...
    Character = 1060577
    Lines = 16365
    Blocks = 1033
    Get b1_matchlist.txt...

    Check...
    Total Matches = 2,064

    The message indicates b1.txt has 1,033 blocks: the 1,032 image blocks plus the last line indicating the total number of matches retrieved. The message �Total Matches = 2,064� indicates that 2,064 matches in b1.txt agrees with those in b1_matchlist.txt.

    3.2.4   Analysis

    There are several variables used to characterize a Verification or Identification.

    For Verification, the common variables are:

    Positive Verification Rate   = 1 - False Rejection Rate

    False Rejection Rate   = 1 - Positive Verification Rate

    Negative Verification Rate   = 1 - False Acceptance Rate

    False Acceptance Rate   = 1 - Positive Verification Rate

    The single characteristic number should be:
     (1 - False Rejection Rate)  * (1 - False Rejection Rate).


    For Identification, the above method can be used. The alternative is:

    Positive Verification Rate = 1 - False Rejection Rate

    False Rejection Rate = 1 - Positive Verification Rate

     Elimination Rate  =  #Retrieved / #Possible Matches

     Hit Ratio = # Correct Match / #Retrieved


    The single characteristic number should be:

     Positive Verification Rate  * Elimination Rate * Hit Ratio
    There are several numbers that will be required for analysis:

    Possible Matches

    Let the total Images in the input file be N; the Possible Matches will be N*N.
    Attrasoft Matches
    The number of retrieved matches is listed in the last line of b1.txt. Go to the end of b1.txt, you will see something like this:
    Total Number of Matches = 14,300
    Actual Match
    This number depends on your problem. It should be the first number in b1_matchlist.txt.
    Attrasoft Found Duplicates
    Click the �Results� button to get this number, as discussed in the last section.
    Now you have all of the numbers, you can make an analysis.

    Positive Identification

    Positive Identification Rate = the results of clicking the �Results� button divided by the first number in file, b1_matchlist.txt.
    Elimination Rate
    The Elimination Rate is 1 minus the number at the end of b1.txt divided by the number of possible matches.
    Hit Ratio
    The Hit Ratio is the number indicated by the �Results� button divided by the number at the end of b1.txt.
    Composite Index
    Finally, an Identification is measured by the multiplication of Positive Identification * Elimination Rate * Hit Ratio.


    3.2.5   Examples
     

    (1) Initialization and Training

    Click the �Initialize� button; you should see:

    Parameters loaded!
    BioFilter loaded!
    �
    Neural Filter loaded!
    BioFilter 2 loaded!
    Neural Filter 2 loaded!

    (2) Select Input File
    Click the �File Input� button, select  �facefinder_b1_matchlist.txt�; wait for a while, and you will see:

    Number of blocks = 1
    Block 0. Length = 1032
    C:\...\data1\00019ba010_960521.jpg
    C:\...\data1\00019bj010_960521.jpg
    C:\...\data1\00029ba010_960521.jpg
    C:\...\data1\00029bj010_960521.jpg�

    (3) Template
    This step is optional if the default images are used. Click the �Templates� button and wait until the computer calms down. You should see each file name in the input file is listed, indicating the speed of converting images into templates.
    (4) Template Matching
    Click the �Identify� button and wait a few seconds until the b1.txt is opened.


    (5) Check the Results

    Click the �Results� button and see the message of 2,064 matches.


    (6) Analysis

    Total Images N = 1,032 = 516 pairs
    Possible Matches N*N = 1,065,024
    Attrasoft Matches = 14,300

    Actual Duplicates = 2,064
    Attrasoft Found Duplicates = 2,064

    Percent Eliminated = 98.7 %
    = 1 � 14,300/1,065,024
    Percent Duplicate Found = 100 %
    = 2,064/2,064

    To Summarize,


    3.2.6   Identify with the Template File

    Your images were already converted into a template file earlier; you can run it without image conversion. Simply save the template file to c:\...\a1.txt and:


    3.3   Verification

    The FaceFinder can also perform image Verification, but it is much slower (matching several images per second, rather than matching several hundred thousand images per second).

    Verification requires multiple images or templates stored for each ID, say 5 images. The example used here has 516 pairs, i.e. each ID only has only one previous stored image.

    The Verification procedure is exactly the same as the Identification procedure except at the end of the Identification procedure; one more step is added.

    FaceFinder-Verification requires the FaceFinder-Identification as a preprocessing step. FaceFinder-Identification will eliminate 98 % - 99 % of the unmatched images and the Verification procedure will take care of the rest. The output of the Identification procedure, b1.txt, will be used as input for the Verification procedure.

    3.3.1   Operation

    After Identification, there is one more step for Verification:

    Entering data

    Click the �File Input� button and select an input file.
    Verification
    Click the �Verify� button.


    3.3.2   Example

    There is one FaceFinder example in this software. The input file is �b1_first10.txt�, which lists the first 10 blocks in the b1.txt.

    The steps are:

    At the end of the Verification procedure, a web page is opened.
    Return