skip to main content
Tipo de recurso Mostra resultados com: Mostra resultados com: Índice

Coarse-grained correspondence-based ancient Sasanian coin classification by fusion of local features and sparse representation-based classifier

Parsa, Seyyedeh-Sahar ; Sourizaei, Mohamad ; Dehshibi, Mohammad Mahdi ; Esmaeilzadeh Shateri, Reza ; Parsaei, Mohammad Reza

Multimedia Tools and Applications, 2017, Vol.76(14), pp.15535-15560 [Periódico revisado por pares]

Texto completo disponível

Citações Citado por
  • Título:
    Coarse-grained correspondence-based ancient Sasanian coin classification by fusion of local features and sparse representation-based classifier
  • Autor: Parsa, Seyyedeh-Sahar ; Sourizaei, Mohamad ; Dehshibi, Mohammad Mahdi ; Esmaeilzadeh Shateri, Reza ; Parsaei, Mohammad Reza
  • Assuntos: Ancient coin classification ; Cultural heritage ; Feature fusion ; Feature selection ; Kernelized sparse representation-based classification
  • É parte de: Multimedia Tools and Applications, 2017, Vol.76(14), pp.15535-15560
  • Descrição: Numismatics sorts out historical aspects of money. Identification and classification of coins, as a part of their duties, need years of experience. This research aims at using the knowledge of numismatics for developing an image-based classification of ancient Sassanian dynasty coins. A straightforward method is to take coins observe and reverse-side motifs into account, just like numismatics does. To this aim, three feature descriptors, Cosine transform, Wavelet transform and Bi-Directional Principal Component Analysis, are separately applied to the extracted motifs’ areas to form the feature space. To cope with the ‘curse of dimensionality’ and increase the ‘discrimination power’, feature space is enriched with spatial information achieved by applying a feature selection method. Indeed, the best feature subset, which maximizes the mutual information between the joint distribution of the selected features and the classification variable, is selected using the minimum Redundancy Maximum Relevance (mRMR) method to a trade-off between thousands of features and a few hundreds of samples. One fold of our contribution dedicates to decrease the over-fitting probability of the learning model by making the Sparse Representation-based Classifier kernelized. We evaluate our method on a dataset of 573 coin images. The experimental results show that our proposed image representation is more discriminative than the competitive ones in which the system achieves a mean classification rate of 96.51 %.
  • Idioma: Inglês

Buscando em bases de dados remotas. Favor aguardar.