![]() ![]() Recently, many deep-learning-based methods have been successfully applied to banknote recognition. We propose that a machine learning-based method is one of the best candidates to accomplish that since it can be retrained whenever an unrecognized number occurs it does not require the redesign of complicated character segmentation and classification methods. It is desirable for a serial number recognition method to be able to learn how to correctly recognize a previously unrecognizable banknote. Although many methods have been proposed and studied, we believe that existing methods need to be improved in terms of performance and learning capability. Several different classification methods have been investigated including support vector machines (SVM), convolutional neural networks (CNN), hybrid CNN-SVM, and the modified quadratic discriminant function (MQDF). The classification step classifies the image of each digit using the features extracted in the previous step. These may include a histogram of oriented gradients (HOG), intensity values, or Gabor features. The feature extraction step attempts to extract the most useful features of a single digit’s image in order to improve character classification. ![]() One may divide this step into the two sub-steps of feature extraction and classification as was done in a previous investigation. ![]() Many methods for character recognition have been investigated. The second step is character recognition in which the segmented characters are identified. ![]()
0 Comments
Leave a Reply. |
AuthorWrite something about yourself. No need to be fancy, just an overview. ArchivesCategories |