Trademarks cassification by Moment Invariant and FuzzyARTMAP

Abstract

The purpose of automated image classification is to facilitate a machine to classify image patterns without human intervention. There are a variety of approaches proposed to perform the task. In our case, the image chosen is that of trademark. Geometric and Zernike Moment techniques are employed to extract a set of patterns in terms of feature vectors from the image. Fuzzy ARTMAP is then utilized to classify the image patterns. In order to test the invariant properties of the feature vectors, trademark images are manipulated into various orientations in the aspect of rotational, translational and size. The classification performance of Fuzzy ARTMAP is evaluated based on cross validation techniques. It is found that Zernike Moments displayed a higher classification accuracy when compared to Geometric Moments.