A distance-weighed Pearson correlation coefficient based self-organizing networks for image classification

The self organizing maps algorithm of Kohonen one of the popular and widely used in several applications. A drawback of the algorithm is that it doesn´t consider the coherence between the input vector and the weight vector of the network´s neurons. Other expansion of Kohonen´s SOM doesn´t consider updating the winning neurons according to the coherence measure between the two quantities. In this paper an algorithm is being proposed as an expansion of the SOM algorithm that considers both the coherence and distance measures to pick the winning codebook vector and updating the codebook vectors of the network accordingly. The new algorithm´s results shows less correlation error between the neuron and won input vectors.