23.10   NeuralNet Filter

23.10.1   Training

NeuralNet/Training Menu Item

Use the “NeuralNet/Training” menu item to train the NeuralNet Filter to learn the key (what image to look for). These commands will first delete all old training and start to train the software from the beginning. After clicking a button, wait one second. If everything is O.K., a message "Training End!" will be printed in the Status Text Area.

NeuralNet/Retraining Menu Item

Use the “NeuralNet/Retraining” menu item to retrain the software to learn additional keys.

If several keys are used for training, the first learning uses training and all the subsequent learning uses retraining. After clicking the button, wait one second. If everything is O.K., a message "Retraining End!" will be printed in the Status Text Area.

23.10.2   Matching

“NeuralNet/Matching/1:1 (Left vs Right)” Menu Item

Use the “NeuralNet/Matching/1:1 (Left vs Right)” menu item to make a 1:1 Matching. 1:1 Matching compares the key with the image displayed in the right picture box. The “Key” button selects the key image. The right image can be selected by the “F”, “>”, “<” buttons. You must train the software first before 1:1 Matching. Note that, next to each image, a score is printed. The larger the number is, the more similarity between the training image(s) and the retrieved images.

“NeuralNet/Matching/1:N (Key vs Source)” Menu Item

Use the "NeuralNet/Matching/1:N (Key vs Source)" menu item to retrieve images in the search source. You must train the software first before searching. If everything is O.K., a message "Retrieval End!" will be printed in the text area, then a file with a list of retrieved images will be opened.

Note that, next to each image, a score is printed. The larger the number is, the more similarity between the training image(s) and the retrieved images.

“NeuralNet/Matching/N:N (Source vs Source)” Menu Item

Use the " NeuralNet/Matching/N:N (Source vs Source)" menu item to match each image in the search-directory against all other images in the directory. After clicking the button, if everything is O.K., a message "Retrieval End!" will be printed in the text area, then a file with a list of retrieved images will be opened. In the output file, each image in the search-directory has a block; the first line in a block is the input, and the rest of the lines in a block are the output.

Note that, next to each image, a score is printed. The larger the number is, the more similarity between the training image(s) and the retrieved images.

23.10.3   Analysis

For the menu items under NeuralNet/Analysis, please see the Unsupervised Filter section. The functions of the menu items are similar.

23.10.4   Results

For the menu items under NeuralNet/Results, please see the Unsupervised Filter section.  The functions of the menu items are similar.

23.10.5   Parameters on the Main Form

“xywh” Text Boxes

Use the “xywh” Text Boxes to set the training segment. The key-segment is specified by 4 integers:  the upper-left corner (x, y) and the length and height (w, h) of the segment. The units for these four variables are pixels. Once the segment specification is successful, a black box will cover the selected area. If the selected area is not what you want, just re-select the area again.

Segment Button

Use the Segment Button to refresh the training segment setting.

23.10.6   Parameters on the Parameter Window

Neural Net Filter Drop Down List

Use the “Neural Net Filter Drop Down List” to set the Neural Net Filter. The default filter is 100x100. All images are scaled down by an integer amount. For example, 640x480 will be scaled down 7 times to 91x68. You need to understand this, so when you translate, or rotate an image, you will make sure no additional scaling factors are introduced.

The search speed crucially depends on the Neural Net Filter. For example, if the 50x50 filter is used, then the underlying neural net size is reduced by a factor of 4, and the neural computation speed will be increased by a factor of 16.

The filters available are:

•    100x100 (Most Accurate)
•    90x90
•    80x80
•    70x70
•    60x60
•    50x50 (Least Accurate)

Let the speed of 100x100 representation be a base, then the overall speed for:

•    90x90 representation is 1 times faster;
•    80x80 representation is 1.6 times faster;
•    70x70 representation is 2.7 times faster;
•    60x60 representation is 5 times faster; and
•    50x50 representation is 10 times faster.

23.10.7   Parameters on the NeuralNet Parameter Window

NeuralNet/Parameter/Symmetry

Use the "Symmetry" button to set the symmetry. The symmetry settings are:

•    No symmetry;
•    Translation symmetry;
•    Rotation symmetry;
•    Scaling symmetry; and
•    Rotation and Scaling symmetry.

The default setting is Translation Symmetry. To set the symmetry, keep clicking the button; the setting will switch from one to the next each time you click the button.

NeuralNet/Parameter/Translation Type Button

Use the “Translation Type” button to select the accuracy of the Translation Symmetry.  The Translation Type settings (and their codes) are:

•    Most Accurate (0);
•    Accurate (1); and
•    Least (2).

To set the Translation Type, keep clicking the “T Type” button; the setting will switch from one to the next each time you click the button. The default setting is 0, the most accurate setting.

NeuralNet/Parameter/Scaling Type Button

Use the “Scaling Type” button to select the accuracy of the Scaling Symmetry.  The Scaling Type settings (and their codes) are:

•    Least Accurate (0);
•    Accurate (1);
•    Accurate (2); and
•    Most Accurate (3).

To set the Scaling Type, keep clicking the “S Type” button; the setting will switch from one to the next each time you click the button. The default setting is 0, the least accurate setting.

NeuralNet/Parameter/Rotation Type Button

Use the R Type (Rotation Type) buttons to set the Rotation Types. The settings are:   

•    360° rotation (0);
•    -5° to 5° rotation (1);
•    -10° to 10° rotation (2);
•    360° rotation, accurate (3);
•    360° rotation, more accurate (4);
•    360° rotation, most accurate (5).

Other settings can be ordered in a Customized Version.

To set the Rotation Type, keep clicking the “R Type” button; the setting will switch from one to the next each time you click the button. The default setting is 360° rotation (0).

NeuralNet/Parameter/Blurring Text Box

Use the Blurring Text Box to control the amount of output. "0%"-Blurring means the exact match. When the "Blurring" is increased, you will get more and more similar images. As the Blurring goes higher, the speed will be slower. The Blurring settings range from 0 – 50. To set the Blurring, enter a number between 0 and 50 to the text box. The default setting is 10.

NeuralNet/Parameter/Sensitivity Text Box

Use the Sensitivity Text Box to adjust search segment size. The Sensitivity parameter ranges from 0 (least sensitive) to 100 (most sensitive).

•    To search small segment(s), use a high sensitivity search.
•    To search large segment(s), use low sensitivity search.
•    The higher the parameter is set, the more results you will get.

To set the Sensitivity, enter a number between 0 and 100 the text box. The default setting is 50.

NeuralNet/Parameter/External Cut (Threshold) Text Box

Use the External Cut Text Box to eliminate those retrieved images with the weights below the External Cut. This parameter is also called Threshold. To set the External Cut, enter a number to the text box. The default setting is 0. In general, it is better to give no answer than a wrong answer. Assume you are searching images and all similar images have weights ranging from 1,000 to 10,000. It is possible that some other images will pop up with weights ranging from 10 to 100. To eliminate these images, you can set the External Cut to 1,000.

NeuralNet/Parameter/Internal Cut Text Box

The Internal Cut plays a similar role as the External Cut. The Internal Cut ranges from 0 to 100, and the External Cut can be any number. To set the Internal Cut, enter a number between 0 and 100 to the text box. The default setting is 100.

NeuralNet/Parameter/Segment Size Button

Use the “Segment Size” button to select the segment size. The default setting is "L Segment".

•    To search Large Segments, use option 0. For example, if a sample segment is one quarter of the sample image, it is a large segment.
•    To search Small Segments, use option 1. If the segment is 1/25 of the sample image, it is a small segment.

To set the segment size, keep clicking the Segment Size button; the setting will switch from one to the next each time you click the button.

Currently, "Small Segment" only supports Translation Symmetry. If you need Rotation and Scaling symmetry, please use "Large Segment".

Additional symmetries can be added very quickly in a Customized Version.

NeuralNet/Parameter/Image Type Button

There are Black-&-White and Color images. For each of them, there are “sum-search”, “maximum-search”, and “average-search”. This generates 6 image types:

•    BW Sum
•    BW Max
•    BW Avg
•    Color Sum
•    Color Max
•    Color Avg

"BW Sum” is like an integration of function f (x).

"BW Max” is like a maximum value of f (x); and

"BW Avg” is the average of the above two.

 

"Color Sum” is like an integration of function f (x).

"Color Max” is like a maximum value of f (x); and

"Color Avg” is the average of the above two.

To set the image type, keep clicking the Image Type button; the setting will switch from one to the next each time you click the Image Type button.

NeuralNet/Parameter/File Display Type Button

Use the “File Display Type” button to set the output file type. The options are text file and html file.

NeuralNet/Parameter/Auto Segment Button

Use the “Auto Segment” button to select a training segment automatically. The training segment can be specified in two ways:

Manual Specification

Automatic Specification

The default is Manual Specification. In this setting the segment will be specified by the four text boxes (x, y, w, h), as we discussed earlier. If you do not want to pick up a training segment, then let the ImageFinder pick up the segment for you by using the Automatic Specification. This parameter has several settings:

NO Auto Segment

Very Large Segment

Very Large Segment

Large Segment

Large Segment

Medium Segment

Medium Segment

NeuralNet/Parameter/Use BioFilter Button

Use the “Use BioFilter” button to determine whether the BioFilter will be used to eliminate search images.

NeuralNet/Parameter/Use NeuralFilter Button

Use the “Use NeuralFilter” button to determine whether the NeuralFilter will be used to eliminate search images.

Copyright (c) 1998 - 2006 Attrasoft, Inc. All rights reserved.

gina@attrasoft.com