Writer Identification in Old Music Manuscripts Using Contour-Hinge Feature and Dimensionality Reduction with an Autoencoder

摘要: Although most of the previous studies in writer identification music scores assumed successful prior staff-line removal, this assumption does not hold when suffer from a certain level degradation or deformation. The impact removal on result such documents is rather vague. In study, we propose novel method that requires no and segmentation. Staff-line virtually achieved without image processing, by dimensionality reduction with an autoencoder Contour-Hinge feature space. experimental wide range manuscripts shows proposed can achieve favourable results removal.