Abstract: Existing work on DocVQA only considers single-page documents. However, in real applications documents are mostly composed of multiple pages that should be processed altogether. In this work, we propose a new multimodal hierarchical method Hi-VT5, that overcomes the limitations of current methods to process long multipage documents. In contrast to previous hierarchical methods that focus on different semantic granularity (He et al., 2021) or different subtasks (Zhou et al., 2022) used in image classification. Our method is a hierarchical transformer architecture where the encoder learns to summarize the most relevant information of every page and then, the decoder uses this summarized representation to generate the final answer, following a bottom-up approach. Moreover, due to the lack of multipage DocVQA datasets, we also introduce MP-DocVQA, an extension of SP-DocVQA where questions are posed over multipage documents instead of single pages. Through extensive experimentation, we demonstrate that Hi-VT5 is able, in a single stage, to answer the questions and provide the page that contains the answer, which can be used as a kind of explainability measure.

Abstract: Facing high error rates and slow recognition speed for full text transcription of unconstrained handwriting images, keyword spotting is a promising alternative to locate specific search terms within scanned document images. We have previously proposed a learning-based method for keyword spotting using character hidden Markov models that showed a high performance when compared with traditional template image matching. In the lexicon-free approach pursued, only the text appearance was taken into account for recognition. In this paper, we integrate character n-gram language models into the spotting system in order to provide an additional language context. On the modern IAM database as well as the historical George Washington database, we demonstrate that character language models significantly improve the spotting performance.

Abstract: From its early stages, the community of Pattern Recognition and Computer Vision has considered the importance of leveraging the structural information when understanding images. Usually, graphs have been proposed as a suitable model to represent this kind of information due to their flexibility and representational power able to codify both, the components, objects, or entities and their pairwise relationship. Even though graphs have been successfully applied to a huge variety of tasks, as a result of their symbolic and relational nature, graphs have always suffered from some limitations compared to statistical approaches. Indeed, some trivial mathematical operations do not have an equivalence in the graph domain. For instance, in the core of many pattern recognition applications, there is a need to compare two objects. This operation, which is trivial when considering feature vectors defined in \(\mathbb{R}^n\), is not properly defined for graphs.

In this thesis, we have investigated the importance of the structural information from two perspectives, the traditional graph-based methods and the new advances on Geometric Deep Learning. On the one hand, we explore the problem of defining a graph representation and how to deal with it on a large scale and noisy scenario. On the other hand, Graph Neural Networks are proposed to first redefine a Graph Edit Distance methodologies as a metric learning problem, and second, to apply them in a real use case scenario for the detection of repetitive patterns which define tables in invoice documents. As experimental framework, we have validated the different methodological contributions in the domain of Document Image Analysis and Recognition.

Abstract: Given a set of graphs, the median graph has been theoretically presented as a useful concept to infer a representative of the set. However, the computation of the median graph is a highly complex task and its practical application has been very limited up to now. In this work we present a new genetic algorithm for the median graph computation. A set of experiments on real data, where none of the existing algorithms for the median graph computation could be applied up to now due to their computational complexity, show that we obtain good approximations of the median graph. Finally, we use the median graph in a real nearest neighbour classification showing that it leaves the box of the only-theoretical concepts and demonstrating, from a practical point of view, that can be a useful tool to represent a set of graphs.

Abstract: Graph representations have been widely used in pattern recognition thanks to their powerful representation formalism and rich theoretical background. A number of error-tolerant graph matching algorithms such as graph edit distance have been proposed for computing a distance between two labelled graphs. However, they typically suffer from a high
computational complexity, which makes it difficult to apply
these matching algorithms in a real scenario. In this paper, we propose an efficient graph distance based on the emerging field of geometric deep learning. Our method employs a message passing neural network to capture the graph structure and learns a metric with a siamese network approach. The performance of the proposed graph distance is validated in two application cases, graph classification and graph retrieval of handwritten words, and shows a promising performance when compared with
(approximate) graph edit distance benchmarks.