Abstract: The representation (or feature) learning has been an emerging concept in the last years, since it collects a set of techniques that are present in any theoretical or practical methodology referring to artificial intelligence. In computer vision, a very common representation has adopted the form of the well-known Bag of Visual Words. This representation appears implicitly in most approaches where images are described, and is also present in a huge number of areas and domains: image content retrieval, pedestrian detection, human-computer interaction, surveillance, e-health, and social computing, amongst others. The early stages of this dissertation provide an approach for learning visual representations inside evolutionary algorithms, which consists of evolving weighting schemes to improve the BoVW representations for the task of recognizing categories of videos and images. Thus, we demonstrate the applicability of the most common weighting schemes, which are often used in text mining but are less frequently found in computer vision tasks. Beyond learning these visual representations, we provide an approach based on fusion strategies for learning spatiotemporal representations, from multimodal data obtained by depth sensors. Besides, we specially aim at the evolutionary and dynamic modelling, where the temporal factor is present in the nature of the data, such as video sequences of gestures and actions. Indeed, we explore the effects of probabilistic modelling for those approaches based on dynamic programming, so as to handle the temporal deformation and variance amongst video sequences of different categories. Finally, we integrate dynamic programming and generative models into an evolutionary computation framework, with the aim of learning Bags of SubGestures (BoSG) representations and hence to improve the generalization capability of standard gesture recognition approaches. The results obtained in the experimentation demonstrate, first, that evolutionary algorithms are useful for improving the representation of BoVW approaches in several datasets for recognizing categories in still images and video sequences. On the other hand, our experimentation reveals that both, the use of dynamic programming and generative models to align video sequences, and the representations obtained from applying fusion strategies in multimodal data, entail an enhancement on the performance when recognizing some gesture categories. Furthermore, the combination of evolutionary algorithms with models based on dynamic programming and generative approaches results, when aiming at the classification of video categories on large video datasets, in a considerable improvement over standard gesture and action recognition approaches. Finally, we demonstrate the applications of these representations in several domains for human analysis: classification of images where humans may be present, action and gesture recognition for general applications, and in particular for conversational settings within the field of restorative justice

Abstract: Radiomics is an emerging technique for the quantification of imaging data that has recently shown great promise for deeper phenotyping of cardiovascular disease. Thus far, the technique has been mostly applied in single-centre studies. However, one of the main difficulties in multi-centre imaging studies is the inherent variability of image characteristics due to centre differences. In this paper, a comprehensive analysis of radiomics variability under several image- and feature-based normalisation techniques was conducted using a multi-centre cardiovascular magnetic resonance dataset. 218 subjects divided into healthy (n = 112) and hypertrophic cardiomyopathy (n = 106, HCM) groups from five different centres were considered. First and second order texture radiomic features were extracted from three regions of interest, namely the left and right ventricular cavities and the left ventricular myocardium. Two methods were used to assess features’ variability. First, feature distributions were compared across centres to obtain a distribution similarity index. Second, two classification tasks were proposed to assess: (1) the amount of centre-related information encoded in normalised features (centre identification) and (2) the generalisation ability for a classification model when trained on these features (healthy versus HCM classification). The results showed that the feature-based harmonisation technique ComBat is able to remove the variability introduced by centre information from radiomic features, at the expense of slightly degrading classification performance. Piecewise linear histogram matching normalisation gave features with greater generalisation ability for classification ( balanced accuracy in between 0.78 ± 0.08 and 0.79 ± 0.09). Models trained with features from images without normalisation showed the worst performance overall ( balanced accuracy in between 0.45 ± 0.28 and 0.60 ± 0.22). In conclusion, centre-related information removal did not imply good generalisation ability for classification.

Abstract: Pedestrian detection for autonomous driving has gained a lot of prominence during the last few years. Besides the fact that it is one of the hardest tasks within computer vision, it involves huge computational costs. The real-time constraints in the field are tight, and regular processors are not able to handle the workload obtaining an acceptable ratio of frames per second (fps). Moreover, multiple cameras are required to obtain accurate results, so the need to speed up the process is even higher. Taking the work in [1] as our baseline, we propose a CUDA implementation of a pedestrian detection system. Further, we introduce significant algorithmic adjustments and optimizations to adapt the problem to the GPU architecture. The aim is to provide a system capable of running in real-time obtaining reliable results.

Abstract: Best Poster AwardOne of the big challenges of today person detectors is the decreasing of the false positive rate. In this paper, we propose a novel framework to customize person detectors in static camera scenarios in order to reduce this rate. This scheme includes background modeling for subtraction based on gradient histograms and Mean-Shift clustering. Our experiments show that the detection improved compared to using only the output from the pedestrian detector reducing 87% of the false positives and therefore the overall precision of the detection
was increased signicantly.

Abstract: Historical records of daily activities provide intriguing insights into the life of our ancestors, useful for demography studies and genealogical research. Automatic processing of historical documents, however, has mostly been focused on single works of literature and less on social records, which tend to have a distinct layout, structure, and vocabulary. Such information is usually collected by expert demographers that devote a lot of time to manually transcribe them. This paper presents a new database, compiled from a marriage license books collection, to support research in automatic handwriting recognition for historical documents containing social records. Marriage license books are documents that were used for centuries by ecclesiastical institutions to register marriage licenses. Books from this collection are handwritten and span nearly half a millennium until the beginning of the 20th century. In addition, a study is presented about the capability of state-of-the-art handwritten text recognition systems, when applied to the presented database. Baseline results are reported for reference in future studies.

Abstract: Handwritten marriage licenses books have been used for centuries by ecclesiastical and secular institutions to register marriages. The information contained in these historical documents is useful for demography studies and
genealogical research, among others. Despite the generally simple structure of the text in these documents, automatic transcription and semantic information extraction is difficult due to the distinct and evolutionary vocabulary, which is composed mainly of proper names that change along the time. In previous
works we studied the use of category-based language models to both improve the automatic transcription accuracy and make easier the extraction of semantic information. Here we analyze the main causes of the semantic errors observed in previous results and apply a Grammatical Inference technique known as MGGI to improve the semantic accuracy of the language model obtained. Using this language model, full handwritten text recognition experiments have been carried out, with results supporting the interest of the proposed approach.

Abstract: In this work, we improve the generative replay in a continual learning setting. We notice that in VAE-based generative replay, the generated features are quite far from the original ones when mapped to the latent space. Therefore, we propose modifications that allow the model to learn and generate complex data. More specifically, we incorporate the distillation in latent space between the current and previous models to reduce feature drift. Additionally, a latent matching for the reconstruction and original data is proposed to improve generated features alignment. Further, based on the observation that the reconstructions are better for preserving knowledge, we add the cycling of generations through the previously trained model to make them closer to the original data. Our method outperforms other generative replay methods in various scenarios.

Abstract: Multi-label image classification is about predicting a set of class labels that can be considered as orderless sequential data. Transformers process the sequential data as a whole, therefore they are inherently good at set prediction. The first vision-based transformer model, which was proposed for the object detection task introduced the concept of object queries. Object queries are learnable positional encodings that are used by attention modules in decoder layers to decode the object classes or bounding boxes using the region of interests in an image. However, inputting the same set of object queries to different decoder layers hinders the training: it results in lower performance and delays convergence. In this paper, we propose the usage of primal object queries that are only provided at the start of the transformer decoder stack. In addition, we improve the mixup technique proposed for multi-label classification. The proposed transformer model with primal object queries improves the state-of-the-art class wise F1 metric by 2.1% and 1.8%; and speeds up the convergence by 79.0% and 38.6% on MS-COCO and NUS-WIDE datasets respectively.