Abstract: Historical ciphered manuscripts are a type of document where graphical symbols are used to encrypt their content instead of regular text. Nowadays, expert transcriptions can be found in libraries alongside the corresponding manuscript images. However, those transcriptions are not aligned, so these are barely usable for training deep learning-based recognition methods. To solve this issue, we propose a method to align each symbol in the transcript of an image with its visual representation by using an attention-based Sequence to Sequence (Seq2Seq) model. The core idea is that, by learning to recognise symbols sequence within a cipher line image, the model also identifies their position implicitly through an attention mechanism. Thus, the resulting symbol segmentation can be later used for training algorithms. The experimental evaluation shows that this method is promising, especially taking into account the small size of the cipher dataset.

Abstract: This paper introduces the DocILE benchmark with the largest dataset of business documents for the tasks of Key Information Localization and Extraction and Line Item Recognition. It contains 6.7k annotated business documents, 100k synthetically generated documents, and nearly 1M unlabeled documents for unsupervised pre-training. The dataset has been built with knowledge of domain- and task-specific aspects, resulting in the following key features: (i) annotations in 55 classes, which surpasses the granularity of previously published key information extraction datasets by a large margin; (ii) Line Item Recognition represents a highly practical information extraction task, where key information has to be assigned to items in a table; (iii) documents come from numerous layouts and the test set includes zero- and few-shot cases as well as layouts commonly seen in the training set. The benchmark comes with several baselines, including RoBERTa, LayoutLMv3 and DETR-based Table Transformer; applied to both tasks of the DocILE benchmark, with results shared in this paper, offering a quick starting point for future work. The dataset, baselines and supplementary material are available at https://github.com/rossumai/docile.

Abstract: Presently, man-machine interface development is a widespread research activity. A system to understand hand drawn architectural drawings in a CAD environment is presented in this paper. To understand a document, we have to identify its building elements and their structural properties. An attributed graph structure is chosen as a symbolic representation of the input document and the patterns to recognize in it. An inexact subgraph isomorphism procedure using relaxation labeling techniques is performed. In this paper we focus on how to speed up the matching. There is a building element, the walls, characterized by a hatching pattern. Using a straight line Hough transform (SLHT)-based method, we recognize this pattern, characterized by parallel straight lines, and remove from the input graph the edges belonging to this pattern. The isomorphism is then applied to the remainder of the input graph. When all the building elements have been recognized, the document is redrawn, correcting the inaccurate strokes obtained from a hand-drawn input.

Abstract: This paper proposes a new approach to spot symbols into graphical documents using sparse representations. More specifically, a dictionary is learned from a training database of local descriptors defined over the documents. Following their sparse representations, interest points sharing similar properties are used to define interest regions. Using an original adaptation of information retrieval techniques, a vector model for interest regions and for a query symbol is built based on its sparsity in a visual vocabulary where the visual words are columns in the learned dictionary. The matching process is performed comparing the similarity between vector models. Evaluation on SESYD datasets demonstrates that our method is promising.