Neural network for colored line detection?

Can I use a Neural Network to find at least 3 pixels (that have RGB or HSV values in a certain range) in a line? I would like to teach the network to accept when and when not accept the line, is possible to do such thing? The lines are scanned by a scann开发者_StackOverflower, when the network goes looking for some colored lines I first look at the paper with the real picture and then tell the network whether it should accept or not.

Yes, you can, but it would be suboptimal.

Assuming you're only looking at a straight line(and not a full 2-dimensional page), you would have to call the neural network for every group of 3 adjacent pixels on the line(i.e. O(N) times).
And even then the task would be simple enough for you to program the solution directly.

A neural network is good at classifying things, not so much at finding things.
I.e. a good use for an NN would be to recognize handwritten letters on a line. A bad use for it would be to recognize where a line is on a page.