Texture Based Coin Recognition Using Multiple Descriptors

This paper presents a new approach for coin image recognition. The purpose of this paper is to find a number of techniques which are used to find the optimum set of features suitable for the coin recognition. Three different approaches are used, the first involving Gabor filter, the second based on Hu invariant moments and the third based on a set of hybrid invariant moments derived from normalized central image moments. Prior to the feature extraction, a pre-processing step is introduced to enhance the content of the image. The features are applied to a dataset of Indian coins divided into four classes and four categories in each class, and classification is done based on minimum difference classifier (MDC), Neural Network (NN) and Neuro Fuzzy Classifier (NFC) between training and testing sets. The performance of the classifiers is measured using confusion matrix.