Object classification using a MLP on a selective tuning model

Abstract

Researchers have argued that an attentional mechanism is required to perform many vision tasks. In this paper we propose an approach to object classification that is based on the multi layer perceptron neural network implemented on the selective tuning model, in order to classify the scan-path of an object. A form of scan-path is obtained using the selective tuning model. The neural network takes as input this scan-path and gives, as output, the estimated class. The entire structure can learn, from a wide variety of examples, how to classify scan-path patterns in a supervised manner and then to recognize objects in digital images. This model of selective visual attention provides for a solution to the problems of selection in an image and information routing through the visual processing hierarchy. This approach is described in some detail and a performance example of scan-path classification is shown. The results confirm that the selective tuning model is both robust and fast