Artificial neural network
Example 2:
Reconstruct 2-dimensional color space after training neural network using 3 points of color value as shown in the following image.
| Point # | X | Y | Red channel (0~255) | Green channel (0~255) | Blue channel (0~255) |
| 1 | 0.5 | 0.1 | 255 | 0 | 0 |
| 2 | 0.1 | 0.9 | 0 | 255 | 0 |
| 3 | 0.9 | 0.9 | 0 | 0 | 255 |
where X and Y mean geometrical location of each point.
For reference, 24 bit color consists of red, green, and blue channels from 0 to 255
1. First of all, setup neural network topology.
Click on 'Artificial neural network' under Menu > Tool
There are 2 input variables such as X, Y and 3 output variables such as Red, Green, Blue.
2. Prepare training pattern
Since every input or output values should have values less than 1, color channel value (0~255) will be normalized to be 1 simply dividing maximum number (255).
For simplicity, a user can use Excel. Make the following table and then copy the table to clipboard!!
3. To import the excel table, click on the 'Training pattern' button and then click on the 'Paste all columns' button. You can see the imported table in the window.
4. Start training by clicking on 'Start training' button. It will take about 20 seconds.
To check regression result, click on the 'Recall' button. You will se the new window titled 'Neural Network - Recalling Pattern'
5. When training is completed, save it by clicking on the 'Save Network' button.
File name is 'Sample 2DColorMap - Trained network.vgn'
6. Develop script code and execute the code
Click on 'Module Library' under Menu>Tool.
Then add new module and finally click on 'Test run' button to execute the code.
VBScript source code
|
Function Main()
'---- Parameters ------------------- ResolutionSize=5 ViewPortSize=350
'---- Define Canvas ------------------------ Call CustomUI.Define_Canvas(ViewPortSize,ViewPortSize) Call CustomUI.Clear_Canvas CustomUI.Form_BringToFront CustomUI.Form_Caption="2D color mapping using artificial neural network"
'--- Load trained neural network map --------------- Call NeuralNet.OpenNN_Once("Sample 2DColorMap - Trained network.vgn")
'---- Draw 2D map ---------------------------------------------- For y= 0 to ViewPortSize step ResolutionSize For x=0 to ViewPortSize step ResolutionSize
'Assign inputs NeuralNet.InputData(1)=x/ViewPortSize NeuralNet.InputData(2)=y/ViewPortSize
'Predict Call NeuralNet.PredictNN()
'Read outputs Red=NeuralNet.OutputData(1)*255 Green=NeuralNet.OutputData(2)*255 Blue=NeuralNet.OutputData(3)*255
'Set fill color and draw rectangle Call CustomUI.Set_FillColor_byRGB(Red,Green,Blue) Call CustomUI.FillRectangle(x,y,ResolutionSize,ResolutionSize) Next Next
'--- Finalize CustomUI ----------------- CustomUI.Make_Clone
End Function |
