Sign in to follow this  

Fitness function for ANN for character recognition

Recommended Posts

Title says it all, but let me explain. I've created some examples of application that use neural networks. And I've came to an application that will analyze the user's handwriting, and convert it to a character. The main problem is developing the fitness function for the artificial neural network. I have the following input: A map (in this case 10x10, but it can be larger, smaller, anything) of values between 0 and 1, this map is the blurred character the user has drawn. I have the same number of outputs as the number of characters the user has drawn for the neural network. For example, if the user has drawn 3 characters, the neural networks contains 3 outputs. A single output can have the values between 0 and 1. I've tried several fitness functions, without much of a success. I've tried the following: - add up all the output values that aren't the character I'm searching for (thus creating a error-value). This makes the output value for the correct character random, so it didn't work. - add up all the output values that aren't the character I'm searching for (again, an error value). Divide this value by the correct character value. Somehow, the values aren't very consistent, and it seems to pick the wrong character very quick. Do any of you have an idea of how to create an algorithm for this? To sum it up, I have the following information: - 10x10 input values between 0 and 1 - x output values, with (x - 1) wrong values, and 1 correct value. The correct value should be the highest of them all. P.S. I'm not sure if this topic should be in math, but I didn't know where it would fit best. Thanks in advance, Max Henger

Share this post

Link to post
Share on other sites

Create an account or sign in to comment

You need to be a member in order to leave a comment

Create an account

Sign up for a new account in our community. It's easy!

Register a new account

Sign in

Already have an account? Sign in here.

Sign In Now

Sign in to follow this