Large scale style based dating of medieval manuscripts

Abstract

In this paper we propose a novel approach for manuscript dating based on shape statistics. Our goal was to develop a strategy well suited for a large scale dating effort where heterogeneous collections of thousands of manuscripts could be automatically processed. The proposed method takes the gray scale image as input, then uses the stroke width transform and a statistical model of the gradient image to find ink boundaries. Finally, a distribution over common shapes, quantified using shape context descriptors, is produced for each manuscript image. The proposed method is binarization-free, rotational invariant and requires minimal segmentation. We evaluate our work on the 10000+ manuscripts collection "Svenskt diplomatariums huvudkartotek", consisting of charters from the medieval period of todays Sweden. The images, originally intended for web viewing, were of low quality and had compression artifacts. Due to unsupervised feature learning and regression, the collection could be dated with a median absolute error below 19 years even though we only used 5% of the labels in the estimator training.

Historical Manuscript Production Date Estimation Using Deep Convolutional Neural Networks

“…Comparing P80 for best and worst estimators is approximately 15 years, which we argue is high. Note that the comparisons from [1] and [2] are produced using a low resolution version of SDHK and hand crafted features (and in the latter case, transcribed text).…”

“…In [1] and [2], we used a GP to do regression over a bag-of-features type feature space for manuscript dating. There, the feature extraction was handcrafted, while in this paper the CNN did the feature learning.…”

“…In our earlier work [1], [2], we have showed that a Gaussian Process (GP), in combination with handcrafted ink stroke descriptors, can be used for manuscript dating. In [3], Support Vector Regression (SVR) with a two phase estimation refinement have been used in combination with handcrafted features.…”

Historical Document Dating Using Unsupervised Attribute Learning

“…In [12], the problem of dating historical color photographs has been studied, using features which capture temporally discriminative information. The dating of medieval manuscripts based on the "Svenskt diplomatariums huvudkartotek" (SDHK) data set has been proposed in [3], with corpus spanning from the years 1050-1523. The date of the document is estimated using regression methods with shape features.…”

“…Automatic date estimation provides an efficient tool for scholars who work with a large digitized database. Therefore, it has attracted several researches [1], [2], [3], [4], [5]. Document representations play an important role in automatic date estimation based on the writing style.…”