Rough Set on Trademark Images for Neural Network Classifier

Rough set theory (RS), introduced by Zdzislaw Pawlak in the early 1980s, is a methodology that concerned with the classificatory analysis of imprecise, uncertain or incomplete information or knowledge expressed in terms of data acquired from experiences or observations. It has the ability to distinguish between object and reason about the objects in the universe in which objects are perceived through the information that is available about them through the values for a predetermined set of attribute. The main advantage of RS is that it requires no additional information to the data represented in table. On the other hand, Supervised Neural Network learns by abstracting a mapping function from the training data for classification purposes. However the drawback of using a supervised neural network is that a large amount of training data must be provided for it to obtain an accurate mapping function. The problem is further aggravated if the data are in the continuous form (real values). Thus, in this paper we overcome the problem by transforming the training data in the continuous form into discrete values using Rough Sets theory and Boolean Reasoning technique. Here, global shape features are chosen to represent the logo images. The invariant features representing logo images are obtained by using the Geometric Invariant Moment Technique (Hu, 1962). The classification results prove that discretization using Rough Sets and Boolean Reasoning can reduce the training cycle and significantly increase the accuracy of the classification of logo images.