| This dissertation describes and tests a computerized model of human visual pattern classification. This model consists of three stages: early visual processing, preattentive processing, and cognitive processing. The first and third stages of this model have previously been studied in great detail. Therefore, the majority of the dissertation concentrated on the examining the second stage of the model which was implemented with an artificial neural network. The computerized neural network model was evaluated by its ability to simulate the data from six studies found in the psychological literature of human pattern classification. All the studies involved the classification of random dot patterns that were first introduced by Posner. The results of the six computer simulations closely matched the human data. The major variables examined were speed of learning, limits of distortion, learning set size, number of categories, delayed testing, and generalization to novel patterns.; After the simulations were completed, all three stages of the model were used to test the model on the real world problem of fingerprint classification. A set of classified fingerprints were obtained from the FBI. These fingerprints were used to train and test the visual pattern classification model. The model was able to learn the training patterns fairly easily, and it performed well on a set of fingerprints that it had never seen. These results give optimism to the belief that this model can be used not only to model human visual properties, but also to help solve longstanding difficult visual classification problems. |
