Musicologist-driven writer identification in early music manuscripts

Abstract

Recent renewed interest in computational writer identification has resulted in an increased number of publications. In relation to historical musicology its application has so far been limited. One of the obstacles seems to be that the clarity of the images from the scans available for computational analysis is often not sufficient. In this paper, the use of the Hinge feature is proposed to avoid segmentation and staff-line removal for effective feature extraction from low quality scans. The use of an auto encoder in Hinge feature space is suggested as an alternative to staff-line removal by image processing, and their performance is compared. The result of the experiment shows an accuracy of 87 % for the dataset containing 84 writers' samples, and superiority of our segmentation and staff-line removal free approach. Practical analysis on Bach's autograph manuscript of the Well-Tempered Clavier II (Additional MS. 35021 in the British Library, London) is also presented and the extensive applicability of our approach is demonstrated.

Towards Musicologist-Driven Mining of Handwritten Scores

“…In our previous paper [7], musicologist-driven writer identification has been investigated, and the proposed algorithm is designed to be stand-alone. This paper concerns a method of extracting numerical evidence with uncertainty to be fused with other evidence.…”

“…Although texture-based features can be extracted from music manuscripts without segmentation [7], subtler analysis such as the specification of chronological order needs feature analysis from smaller symbols such as clefs as explained in the previous section. The extraction of C-clefs from manuscripts requires accurate segmentation.…”

Patch-Based Convolutional Neural Network for the Writer Classification Problem in Music Score Images

“…In [12], a Hidden Markov Model (HMM) was used with a Blurred Shape Model (BSM) descriptor, applied to the CVC-MUSCIMA dataset without staff removal. In [13], Hinge features were used in a small dataset with 88 music score samples. The work presents results with the global analysis of the image.…”

“…In this context, several works regarding WPC [2, 3,4] presented approaches based on the use of hand-designed feature extractors. Some of them were applied to solve problems related to music score documents [12,13]. In contrast, this work proposes the use of a CNN to learn the feature extractor and the classifier jointly, which is an approach that has already given satisfactory results in problems belonging to a wide variety of areas [7] [2] [4].…”

“…In contrast, this work proposes the use of a CNN to learn the feature extractor and the classifier jointly, which is an approach that has already given satisfactory results in problems belonging to a wide variety of areas [7] [2] [4]. Another approach that has been explored in recent works, similar to those presented by [10] and [13], is based on the analysis of local characteristics of images by extracting patches. The patch extraction strategy also presented good results on medical image analysis [14].…”