16.   SegLocator
16.1   Coordinate System
16.2   Six Examples
16.3   Segment Output

16.   SegLocator

In this chapter, we will introduce Segment Locator or SegLocator. The SegLocator is a specialized NeuralNet Filter, which will locate a segment and print their coordinates.
The SegLocator works exactly the same as the ImageFinder; however the SegLocator only supports 1:N Matching in the directory input mode, i.e. the SegLocator does not support file input and long-search modes. The available commands are:

The �SegLocator/Batch Run� only supports batch code 1027, which is a consecutive execution of the following two commands: The SegLocator output is the ImageFinder output plus the coordinates of the segment located in the images.
 

16.1   Coordinate System

The data for this chapter consists of the four images below:


---------

------

------

Figure 16.1 The data consists of 4 images.
Note: Each image contains 9 logos.
 

There are 9 trademarks randomly placed in the four images. We want the SegLocator not only to recognize the image segment, but also locate the position.

O                               x

y 

Figure 16.2 Segment Specification.

The image segment is specified by 5 numbers:

(x, y, w, h, r)

where:

 x is the x-coordinate of the upper left corner;
 y is the y-coordinate of the upper left corner;
 w is the x-length of the segment (width);
 h is the y-length of the segment (height); and
 r is the rotation angle.

Figure 16.3   SegLocator.
 

The default Representation is 100x100. The x coordinate runs from 0 to 100 and the y coordinate runs from 0 to 100. (This depends on the NeuralNet filter selected. If you choose a 50x50 NeuralNet Filter instead of 100x100, then the x coordinate runs from 0 to 50 and the y coordinate runs from 0 to 50.) The origin of the coordinate system is at the upper left corner. The x-axis runs from left to right and the y-axis runs from top to bottom.

To find a segment in an image, you can use:

Note that the image area in the ImageFinder is 300x300. You can find the location of a segment in an image by viewing it in the ImageFinder, but remember in the ImageFinder, the image area is 300x300, while the output specification is 100x100.

16.2   Six Examples

1. The United Way.

You can duplicate this example by two clicks:

If you want to set the parameter manually,

Input

Key Segment = image002
 
NeuralNet Parameter
Training Segment 98 16 98 81


The segment picked is 98 16 98 81. This is in 300x300 scale because the ImageFinder display area is 300 x 300 pixels. The output is 100x100 scale; and (98 16 98 81) in 300x300 scale will convert to (33, 5, 33, 27) in 100x100 scale. The output is:

image002.jpg  86976
 34  7  26  20  0
image004.jpg  86528
 2  4  26  20  0
image006.jpg  84608
 63  33  26  20  0
image008.jpg  82368
 8  10  26  20  0


In all 4 cases, the �United Way� segments are correctly identified.

The output for the first image is (34  7  26  20) and the original segment is (33, 5, 33, 27). The difference is from the fact that the SegLocator locates the image segment without a blank area, while the original training segment, (33, 5, 33, 27), contains a blank area.

2.   Monopoly.

You can duplicate this example by two clicks:

If you want to set the parameter manually,

Input

 Key Segment = image002
 
NeuralNet Parameter
Training Segment 85 117 110 68


The segment picked, in 300x300 scale, is 85 117 110 68, which translates to 100x100 scale as: (28, 39, 37, 23). The output is:

image002.jpg  78784
 31  42  29  8  0
image004.jpg  78336
 35  74  29  8  0
image006.jpg  78080
 34  77  29  8  0
image008.jpg  76480
 7  77  29  8  0


In all 4 cases, the �Monopoly� segments are correctly identified.

The output for the first image is (31  42  29  8) and the original segment is (28, 39, 37, 23). The difference is from the fact that the SegLocator locates the image segment without a blank area, while the original training segment contains a blank area.

3. Mr. Potato.

You can duplicate this example by two clicks:

If you want to set the parameter manually,

Input

Key Segment = image002
 
NeuralNet Parameter
Training Segment is 206 17 74 88
The segment picked, in 300x300 scale, is 206 17 74 88. The output is:
image002.jpg  79552
 73  9  12  20  0
image004.jpg  78080
 15  68  12  20  0
image006.jpg  78336
 40  41  12  20  0
image008.jpg  75584
 48  11  12  20  0
In all 4 cases, the �Mr. Potato� segments are correctly identified.

4.   AAA.

You can duplicate this example by two clicks:

If you want to set the parameter manually,

Input

Key Segment = image002
 
NeuralNet Parameter
Training Segment is  109 194 74 75
The output is:
image002.jpg  79872
 37  66  19  14  0
image004.jpg  79296
 68  41  19  14  0
image006.jpg  77376
 7  12  19  14  0
image008.jpg  76288
 39  43  19  14  0
In all 4 cases, the �AAA� segments are correctly identified.

5.   Ford.

You can duplicate this example by two clicks:

If you want to set the parameter manually,

Input

Key Segment = image002
 
NeuralNet Parameter
Training Segment is  195 202 92 60
The output is:
image002.jpg  78976
 62  70  22  8  0
image004.jpg  78912
 66  14  22  8  0
image006.jpg  77888
 40  16  22  8  0
image008.jpg  75904
 7  47  22  8  0


In all 4 cases, the �Ford� segments are correctly identified.
 

6.   Soup.

You can duplicate this example by two clicks:

If you want to set the parameter manually,

Input

Key Segment = image002
 
Image Processing
Threshold File = Customized filter:
Red 30 - 127 Light Background
Green: Ignore
Blue: Ignore


NeuralNet Parameter

Training Segment is  18 14 53 69
Blurring = 2
Segment Size = S(Small) Segment


The output is:

image002.jpg  2400000
 10  7  8  13  0
image004.jpg  48000
 50  46  8  13  0
image006.jpg  60000
 18  69  8  13  0
image008.jpg  48000
 73  20  8  13  0
In all 4 cases, the �Soup� segments are correctly identified.

16.3   Segment Output

The output of the SegLocator is within an area of 100x100. Interpretation of the SegLocator�s output depends on:

Assume: Under the Integer Reduction Filters, the new width is 256/4 = 64, and the new height is 384/4 = 96. Assume the SegLocator identifies a segment in (x, y, w, h) = (10, 10, 20, 20), then in terms of pixel value, it means 4 * (10, 10, 20, 20) = (40, 40, 80, 80).

Assume:

Under the Real Reduction Filters, the new width is 256/3.8 = 67, and the new height is 384/3.8 = 100. Assume the SegLocator identifies a segment in (x, y, w, h) = (10, 10, 20, 20), then in terms of pixel value, it means 3.8 * (10, 10, 20, 20) = (38, 38, 76, 76).
Return