A Study of Self Organizing in Image Recognition Application

This research focuses on a self-organizing network and the effect of some factors on the network performance by using it for personal photo recognition. The network is built in two types: one dimensional, and two-dimensional network. A study for effect of the types of initial weights on the performance of the network as well, the performance of the network is tested using two types of the initial weights: random weights and fixed weights, which represent the mean of the inputs to the network. The obtained results show that it is preferred to initialize the network with random weights for the one dimensional network while the results were almost equal for the two dimensional network. The research also studies the effects of the neighborhood functions on the performance of the network.
Three neighborhood functions are applied and compared. The experimental results had proved that the second function is the most efficient function among the others. In addition, a study of the network acceptance for the corrupted photos, and the effect of the number of output nodes on the ability of the network recognition. The results show that the density of the weight vectors should be greater than the network to separate the training patterns in the output nodes..