17.   Customized Point-Locator
17.1   Locating Eye and Nose
17.2   Feret 100
17. 3   Run Your Data
17.4   Analysis


17.   Customized Point-Locator

The Point-Locator is a special version of the Segment-Locator; it locates points by locating the segment first and specifies points based on the position of the located segment.

This chapter introduces a Customized Point-Locator, in which the training segment is already designed and encoded into the software. You do not need to train the software. You will run the software with the “Batch/Run” command.

17.1   Locating Eye and Nose

You can run this example with three clicks:


 

Figure 17.1 Customized Point-Locator.

Figure 17.2 Visual results of Customized Point-Locator.
 

In this example, we will locate the eyes and nose in a face. This customized module will work only for the following images:

1. Ferret Database Face Image, 256x384 pixels;
2. Only front view images can be used.

The first 20 images in the Feret database will be used in this example.

We will now explain this example. The first click, “Example/SegLocator/Point Locator: Feret 20”, will load the batch file. The second click, “Batch/Run”, will execute this batch file. Note  the execution code 1101. Each customized version has its own batch code. The Point-Locator will:

1. Locate the Face;
2. Locate the left eye, the right eye, and nose ROI (Region Of Interest);

After the Point-Locator locates the left eye segment, the right eye segment, and the nose segment, it will compute their coordinates based on the located segments. The training images are hard coded into the Customized Point-Locator, therefore, there is no need for training. The results will look like this:

0. 00007fa010_930831.jpg

 Left eye: 98.0  165.0
 Right: 159.0  161.0
 Nose: 127  199
1. 00002fb010_930831.jpg
 Left eye: 107.0  166.0
 Right: 158.0  157.0
 Nose: 126  184
2. 00003fb010_941121.jpg
 Left eye: 89.0  192.0
 Right: 153.0  196.0
 Nose: 119  229
…

The third click, “Example/SegLocator/Point Locator: Results”, will show the results visually; the Point-Locator will draw a 40x30 square on left eye, the right eye, and the nose. The Point-Locator will present an animation, which runs through all images. This will last 80 seconds, 4 seconds per image.

17.2   Feret 100

You can run this example with three clicks:

This example will use images in the “.\Feret100\” folder. There are 100 images in this folder. Other than the numbers, this example is similar to the previous example.
 

17. 3   Run Your Data

You can use other Feret images by using a different folder. Here is the operation:

17.4   Analysis

Now we will compare the output of the Point-Locator for the “Feret 20” example with human work. The following are two tables for the left eye and the right eye. The images are 256x384 pixels in size. The units for all measurements are pixels. Each table has the following columns:
 

Image Number
Human Result: x1
Human Result: y1
Point-Locator: x2
Point-Locator: y2
Gap = |x2-x1| + |y2-y1|


Left Eye:

          x1      y1       x2          y2
1        98    165       101.0     164.0       4.0
2       106   165       107.0     166.0       2.0
3       88     191       89.0       192.0       2.0
4       92     171       91.0       172.0       2.0
5       102   150       103.0     150.0       1.0
6       112   175       113.0     177.0       3.0
7       95     163       98.0       165.0       5.0
8       91     170       93.0       171.0        3.0
9       106   179       107.0     180.0        2.0
10      124   185      126.0     188.0        5.0
11      112   158      112.0     157.0        1.0
12*    112   159      112.0     171.0        12.0
13      112   180      110.0     181.0        3.0
14      121   170      121.0     165.0        5.0
15      107   184      108.0     185.0        2.0
16      104  164       105.0     166.0        3.0
17      123   183      123.0     184.0        1.0
18      96     176      94.0       177.0        3.0
19      104   158      106.0     160.0        4.0
20      119   150      120.0     151.0        2.0
sum      65.0

Here * means error. There are 40 measurements total (20 pairs of (x,y)). The total gap is 65, and the average measurement Gap = 65/40= 1.6. There is one error; Error Rate = 2.5% = 1/40.
 

Right Eye:

           x1      y1        x2        y2

1*      161      163     149.0      165.0  14.0
2       158      160     158.0      157.0  3.0
3       152      194     153.0      196.0  3.0
4       149      167     147.0      167.0  2.0
5       158      145     159.0      145.0  1.0
6       170      173     168.0      173.0  2.0
7       160      159     159.0      161.0  3.0
8       148      161     148.0      162.0  1.0
9       160      173     160.0      174.0  1.0
10      176      186     178.0      188.0  5.0
11      168      160     164.0      161.0  5.0
12      169      171     171.0      173.0  4.0
13      168      181     171.0      183.0  5.0
14      181      174     178.0      174.0  3.0
15      160      180     158.0      180.0  2.0
16      150      163     151.0      164.0  2.0
17      177      187     179.0      188.0  3.0
18      142      176     143.0      176.0  1.0
19      165      162     167.0      163.0  3.0
20      161      147     162.0      149.0  3.0
sum     65.0
 

Average measurement Gap = 65/40 =1.6
Error Rate = 2.5% = 1/40.

Other than a few percentages of errors, the human reading and the Point-Locator readings are very close.
 
 

Return