How do I create training data set for deep learning?

I am trying to create a convolutional neural network from a unsupervised learning.

The training data set I want to create contains 10 examples (x_i, y_i) , i=1,2,...10 .

the x_i and y__i are both certain types of images.

So when using the convolutional neural network, I would like to input x_i and train the output as y_i.

Is it possible to create this training data set on matlab, and if so, how can I do that?

[KSSV]

What exactly are the images? What do you want to train? What is input and what you are targetting?

[Casey]

@KSSV Thank you for your comment.

My images are propagated light beam wavefront's intensity distribution(x) and phase distribution(y), which involves the atmospheric turbulence.

So the input would be the intensity distribution, and through the training of CNN, I would like to predict its phase distribution.