Page 204 - DCAP208_Management Support Systems

Unit 12: Applications of Neural Network

Now train the neural network with the error function: Notes

Where,

nx, ny are the dimensions (number of pixels) of the image in the x, y directions
Each pixel is specified by x , y, z where z in the gray intensity of the pixel
i j ij ij
o(x ,y) is the output of the neural network when the input is the position of the pixel i, j.
i j
More complex images will require larger architectures.
Check the results of the trained neural network.

Compare original (left) with NN output (right):
Figure 12.6: Comparing Original Image with NN Output

Source: http://wwwteor.mi.infn.it/~rojo/teaching/Milano-NNet-Course-Lecture4.pdf
Using Image Recognition

Once the NN has been trained, we can use it for image recognition. This can be done by comparing
the error function between the trained ANN and another different image → Below a given
threshold (to be validated) the two images are assigned to be identical.

!
Caution The trained ANN works also as compressed version of the image, where all the
information of the image is now encoded into the weights and thresholds.

12.2.2 Evolution of Neural Networks for Control of Pursuit & Evasion

The following MPEG movie sequences illustrate behaviour generated by dynamical recurrent
neural network controllers co-evolved for pursuit and evasion capabilities. From an initial
population of random network designs, successful designs in each generation are selected for
reproduction with recombination, mutation, and gene duplication. Selection is based on measures
of how well each controller performs in a number of pursuit-evasion contests. In each contest a
pursuer controller and an evader controller are pitched against each other, controlling simple

LOVELY PROFESSIONAL UNIVERSITY 197