Abstract: Document Visual Question Answering (DocVQA) refers to the task of answering questions from document images. Existing work on DocVQA only considers single-page documents. However, in real scenarios documents are mostly composed of multiple pages that should be processed altogether. In this work we extend DocVQA to the multi-page scenario. For that, we first create a new dataset, MP-DocVQA, where questions are posed over multi-page documents instead of single pages. Second, we propose a new hierarchical method, Hi-VT5, based on the T5 architecture, that overcomes the limitations of current methods to process long multi-page documents. The proposed method is based on a hierarchical transformer architecture where the encoder summarizes the most relevant information of every page and then, the decoder takes this summarized information to generate the final answer. Through extensive experimentation, we demonstrate that our method is able, in a single stage, to answer the questions and provide the page that contains the relevant information to find the answer, which can be used as a kind of explainability measure.

Abstract: Tabular structures in business documents offer a complementary dimension to the raw textual data. For instance, there is information about the relationships among pieces of information. Nowadays, digital mailroom applications have become a key service for workflow automation. Therefore, the detection and interpretation of tables is crucial. With the recent advances in information extraction, table detection and recognition has gained interest in document image analysis, in particular, with the absence of rule lines and unknown information about rows and columns. However, business documents usually contain sensitive contents limiting the amount of public benchmarking datasets. In this paper, we propose a graph-based approach for detecting tables in document images which do not require the raw content of the document. Hence, the sensitive content can be previously removed and, instead of using the raw image or textual content, we propose a purely structural approach to keep sensitive data anonymous. Our framework uses graph neural networks (GNNs) to describe the local repetitive structures that constitute a table. In particular, our main application domain are business documents. We have carefully validated our approach in two invoice datasets and a modern document benchmark. Our experiments demonstrate that tables can be detected by purely structural approaches.

Abstract: Change detection is the most important task for video surveillance analytics such as foreground and anomaly detection. Current foreground detectors learn models from annotated images since the goal is to generate a robust foreground model able to detect changes in all possible scenarios. Unfortunately, manual labelling is very expensive. Most advanced supervised learning techniques based on generic object detection datasets currently exhibit very poor performance when applied to surveillance datasets because of the unconstrained nature of such environments in terms of types and appearances of objects. In this paper, we take advantage of change detection for training multiple foreground detectors in an unsupervised manner. We use statistical learning techniques which exploit the use of latent parameters for selecting the best foreground model parameters for a given scenario. In essence, the main novelty of our proposed approach is to combine the where (motion segmentation) and what (learning procedure) in change detection in an unsupervised way for improving the specificity and generalization power of foreground detectors at the same time. We propose a framework based on latent support vector machines that, given a noisy initialization based on motion cues, learns the correct position, aspect ratio, and appearance of all moving objects in a particular scene. Specificity is achieved by learning the particular change detections of a given scenario, and generalization is guaranteed since our method can be applied to any possible scene and foreground object, as demonstrated in the experimental results outperforming the state-of-the-art.

Abstract: In this paper we propose an unsupervised segmentation-free method for word spotting in document images. Documents are represented with a grid of HOG descriptors, and a sliding-window approach is used to locate the document regions that are most similar to the query. We use the Exemplar SVM framework to produce a better representation of the query in an unsupervised way. Then, we use a more discriminative representation based on Fisher Vector to rerank the best regions retrieved, and the most promising ones are used to expand the Exemplar SVM training set and improve the query representation. Finally, the document descriptors are precomputed and compressed with Product Quantization. This offers two advantages: first, a large number of documents can be kept in RAM memory at the same time. Second, the sliding window becomes significantly faster since distances between quantized HOG descriptors can be precomputed. Our results significantly outperform other segmentation-free methods in the literature, both in accuracy and in speed and memory usage.

Abstract: In this work, a new one-class classification ensemble strategy called approximate polytope ensemble is presented. The main contribution of the paper is threefold. First, the geometrical concept of convex hull is used to define the boundary of the target class defining the problem. Expansions and contractions of this geometrical structure are introduced in order to avoid over-fitting. Second, the decision whether a point belongs to the convex hull model in high dimensional spaces is approximated by means of random projections and an ensemble decision process. Finally, a tiling strategy is proposed in order to model non-convex structures. Experimental results show that the proposed strategy is significantly better than state of the art one-class classification methods on over 200 datasets.

Abstract: The automatic transcription of unconstrained continuous handwritten text requires well trained recognition systems. The semi-supervised paradigm introduces the concept of not only using labeled data but also unlabeled data in the learning process. Unlabeled data can be gathered at little or not cost. Hence it has the potential to reduce the need for labeling training data, a tedious and costly process. Given a weak initial recognizer trained on labeled data, self-training can be used to recognize unlabeled data and add words that were recognized with high confidence to the training set for re-training. This process is not trivial and requires great care as far as selecting the elements that are to be added to the training set is concerned. In this paper, we propose to use a bidirectional long short-term memory neural network handwritten recognition system for keyword spotting in order to select new elements. A set of experiments shows the high potential of self-training for bootstrapping handwriting recognition systems, both for modern and historical handwritings, and demonstrate the benefits of using keyword spotting over previously published self-training schemes.

Abstract: Genetic Algorithms (GA) have been previously applied to Error-Correcting Output Codes (ECOC) in state-of-the-art works in order to find a suitable coding matrix. Nevertheless, none of the presented techniques directly take into account the properties of the ECOC matrix. As a result the considered search space is unnecessarily large. In this paper, a novel Genetic strategy to optimize the ECOC coding step is presented. This novel strategy redefines the usual crossover and mutation operators in order to take into account the theoretical properties of the ECOC framework. Thus, it reduces the search space and lets the algorithm to converge faster. In addition, a novel operator that is able to enlarge the code in a smart way is introduced. The novel methodology is tested on several UCI datasets and four challenging computer vision problems. Furthermore, the analysis of the results done in terms of performance, code length and number of Support Vectors shows that the optimization process is able to find very efficient codes, in terms of the trade-off between classification performance and the number of classifiers. Finally, classification performance per dichotomizer results shows that the novel proposal is able to obtain similar or even better results while defining a more compact number of dichotomies and SVs compared to state-of-the-art approaches.