Hello!
I looked at the picture you uploaded and it's a hard problem, the fact that some characters overlap makes segmentation difficult. It might be possible to solve it using a convolutional neural network, that would be significantly more time-consuming.
If it's possible to control the input, as to make it easier for a computer to read, I can provide a solution in a few days. What I would suggest is to separate each digit into a box, then we can achieve pretty good accuracy cheaply.
Either way a few thousand examples are required for the neural network to learn to recognize each character.
Please accept my bid only if you can work with a controlled input as I suggested and provide a set of training examples with at least a few thousand occurrences of each character.