Learning Color Constancy Using Neural Networks

2 Learning Color Constancy Using Neural Networks 

A simple artificial neuron simply adds up the signals it receives from other neurons to which it is connected. Each input signal is multiplied by the weight of the connection. Let oj be the output of neuron j and let wt j be the weight between neurons i and j, then the activation a,- of neuron i is computed as 

The activation a,- can be used to compute an output value a, for the neuron i. A threshold is used to define whether the neuron is active or not. Let Oj be the threshold of neuron i, then the output is given by 

Instead of computing a binary output, a sigmoidal function is frequently used to compute the output value. 

The method of Oja (1982) is used to determine the principal axis. Let c be the original data points of the image. Let w = [uy, wg, i/y] be the 3 element weight vector of the neural system. The following update equations are iterated until convergence  

---