Abstract: Sectors as fintech, legaltech or insurance process an inflow of millions of forms, invoices, id documents, claims or similar every day. Together with these, historical archives provide gigantic amounts of digitized documents containing useful information that needs to be stored in machine encoded text with a meaningful structure. This procedure, known as information extraction (IE) comprises the steps of localizing and recognizing text, identifying named entities contained in it and optionally finding relationships among its elements. In this work we explore multi-task neural models at image and graph level to solve all steps in a unified way. While doing so we find benefits and limitations of these end-to-end approaches in comparison with sequential separate methods. More specifically, we first propose a method to produce textual as well as semantic labels with a unified model from handwritten text line images. We do so with the use of a convolutional recurrent neural model trained with connectionist temporal classification to predict the textual as well as semantic information encoded in the images. Secondly, motivated by the success of this approach we investigate the unification of the localization and recognition tasks of handwritten text in full pages with an end-to-end model, observing benefits in doing so. Having two models that tackle information extraction subsequent task pairs in an end-to-end to end manner, we lastly contribute with a method to put them all together in a single neural network to solve the whole information extraction pipeline in a unified way. Doing so we observe some benefits and some limitations in the approach, suggesting that in certain cases it is beneficial to train specialized models that excel at a single challenging task of the information extraction process, as it can be the recognition of named entities or the extraction of relationships between them. For this reason we lastly study the use of the recently arrived graph neural network architectures for the semantic tasks of the information extraction process, which are recognition of named entities and relation extraction, achieving promising results on the relation extraction part.

Abstract: Mobile document image acquisition integrates many distortions which must be corrected or detected on the device, before the document becomes unavailable or paying data transmission fees. In this paper, we propose a system to correct perspective and illumination issues, and estimate the sharpness of the image for OCR recognition. The correction step relies on fast and accurate border detection followed by illumination normalization. Its evaluation on a private dataset shows a clear improvement on OCR accuracy. The quality assessment
step relies on a combination of focus measures. Its evaluation on a public dataset shows that this simple method compares well to state of the art, learning-based methods which cannot be embedded on a mobile, and outperforms metric-based methods.

Abstract: Architectural floor plans exhibit a large variability in notation. Therefore, segmenting and identifying the elements of any kind of plan becomes a challenging task for approaches based on grouping structural primitives obtained by vectorization. Recently, a patch-based segmentation method working at pixel level and relying on the construction of a visual vocabulary has been proposed in [1], showing its adaptability to different notations by automatically learning the visual appearance of the elements in each different notation. This paper presents an evolution of that previous work, after analyzing and testing several alternatives for each of the different steps of the method: Firstly, an automatic plan-size normalization process is done. Secondly we evaluate different features to obtain the description of every patch. Thirdly, we train an SVM classifier to obtain the category of every patch instead of constructing a visual vocabulary. These variations of the method have been tested for wall detection on two datasets of architectural floor plans with different notations. After studying in deep each of the steps in the process pipeline, we are able to find the best system configuration, which highly outperforms the results on wall segmentation obtained by the original paper.