The Study Of Currency Recognition Based On NMF And ENN

Actually as a question of pattern recognition, the research of currency recognition mainly focuses on three aspects:data acquisition, feature extraction and classifier design. Based on the extensive reading of relative literature home and abroad, the Non-negative Matrix Factorization(NMF) is proposed in this paper to extract the characteristics of currency recognition while the Ensemble Neural Network(ENN) is used for the Classifier design.NMF is a kind of novel characteristics extraction method for pattern recognition using part to denote the entirety. The matrix V consisted of currency sample is decomposed by NMF, and the base matrix W can be achieved to express the local feature information while the column vector of H can be understood as the weight coded information of samples on corresponding local characteristics. Compared with conventional methods, the feature information decomposed by NMF coincide in the intuitive experience of human vision better and have more explicit physical meaning.The individual neural networks(NNs) of Ensemble Neural Network(ENN) used for the classifier design is trained via negative correlation learning(NCL). The objective of using negative correlation learning(NCL) is to expertise the individuals on different parts or portion of input patterns in an ensemble making the ensemble learn the whole training data better.The notes image under UV-light is used in thesis as experimental samples and is acted as input vector of the classifiers after a series of preprocessing. The experiment result shows the good performance of the ensemble network based on NCL on the classification of currency recognition and has less misclassification rate compared with the system using individual neutral network and the ensemble network training individual network independently.