LDA based paper currency recognition system using edge histogram descriptor

Paper currency recognition is an important concern for automation to improve our daily monetary activities. Such recognition system uses the banknote images to train a classifier for identification of unknown input notes. One of the basic problems of such system is high dimensional representation of the feature vector (more than 100 dimensions) of note images. Moreover, most of the traditional approaches do not consider to minimize intra-class scatter and maximize inter-class scatter. To get rid of these basic shortcomings, in this paper, we propose an LDA based paper currency recognition method using edge histogram descriptor (EHD). Applying this method, we succeed to represent a note image by a very low dimensional feature vector (around only 15 dimensions). Besides adjusting the scatter of different classes, this method has the ability to tolerate noise of a certain level. We have performed different experiments to support all attractive features of the proposed system. For those experiments, we have used banknotes of different countries and achieved high accuracy with low dimensional feature vector.