Electronic Letters on Computer Vision and Image Analysis (ELCVIA - Universitat Autònoma de Barcelona)
Not a member yet
343 research outputs found
Sort by
Relational Models for Visual Understanding of Graphical Documents.\\ Application to Architectural Drawings.
Full text linkIn this thesis we face the graphical document understanding problem by proposing several relational models for the complete interpretation of floor plans. Firstly, we introduce three different strategies on symbol detection for walls, doors and windows. Secondly, we present two relational strategies that tackle the problem of the visual context extraction. Finally, we construct a knowledge-based model consisting of an ontological definition of the domain and real data. All the resources used in this thesis are freely available for research purposes
Illumination Inconsistency Sleuthing for Exposing Fauxtography and Uncovering Composition Telltales in Digital Images
Full text linkThink about how capture device’s technology is improved day after day. Add to this condition that digital image manipulation tools are increasingly powerful and simple to use. Finally, when a malicious user is added at this equation, the result is an astonishing number of digital images forgeries spread out in the internet as fast as possible. This work present four different methods to fight against image splicing, an special kind of image forgery. Each one of the methods, present significant contributions to state of the art.
Higher-order regularization and morphological techniques for image segmentation
Full text linkImage segmentation is an important field in computer vision and one of its most active research areas, with applications in image understanding, object detection, face recognition, video surveillance or medical image processing. Image segmentation is a challenging problem in general, but especially in the biological and medical image fields, where the imaging techniques usually produce cluttered and noisy images and near-perfect accuracy is required in many cases.In this thesis we first review and compare some standard techniques widely used for medical image segmentation. These techniques use pixel-wise classifiers and introduce weak pair-wise regularization which is insufficient in many cases. We study their difficulties to capture higher-level structural information about the objects to segment. This deficiency leads to many erroneous detections, ragged boundaries, incorrect topological configurations and wrong shapes. To deal with these problems, we propose a new regularization method that learns shape and topological information from training data in a non-parametric way using higher-order potentials
Arabic/Latin and Machine-printed/Handwritten Word Discrimination using HOG-based Shape Descriptor
Full text linkIn this paper, we present an approach for Arabic and Latin script and its type identification based on Histogram of Oriented Gradients (HOG) descriptors. HOGs are first applied at word level based on writing orientation analysis. Then, they are extended to word image partitions to capture fine and discriminative details. Pyramid HOG are also used to study their effects on different observation levels of the image. Finally, co-occurrence matrices of HOG are performed to consider spatial information between pairs of pixels which is not taken into account in basic HOG. A genetic algorithm is applied to select the potential informative features combinations which maximizes the classification accuracy. The output is a relatively short descriptor that provides an effective input to a Bayes-based classifier. Experimental results on a set of words, extracted from standard databases, show that our identification system is robust and provides good word script and type identification: 99.07% of words are correctly classified
Contributions to metric-topological localization and mapping in mobile robotics
Full text linkThis thesis addresses the problem of localization and mapping in mobile robotics. The ability of a robot to build a map of an unknown environment from sensory information is required to perform self-localization and autonomous navigation, as a necessary condition to carry out more complex tasks. This problem has been widely investigated in the last decades, but the solutions presented have still important limitations, mainly to cope with large scale and dynamic environments, and to work in a wider range of conditions and scenarios. In this context, this thesis takes a step forward towards highly efficient localization and mapping.A first contribution of this work is a new mapping strategy that presents two key features: the lightweight representation of world metric information, and the organization of this metric map into a topological structure that allows efficient localization and map optimization. Regarding the first issue, a map is proposed based on planar patches which are extracted from range or RGB-D images. This plane-based map (PbMap) is particularly well suited for indoor scenarios, and has the advantage of being a very compact and still a descriptive representation which is useful to perform real-time place recognition and loop closure. These operations are based on matching planar features taking into account their geometric relationships. On the other hand, the abstraction of metric information is necessary to deal with large scale SLAM and with navigation in complex environments. For that, we propose to structure the map in a metric-topological structure which is dynamically organized upon the sensor observations. Also, a simultaneous localization and mapping (SLAM) system employing an omnidirectional RGB-D device which combines several structured-light sensors (Asus Xtion Pro Live) is presented. This device allows the quick construction of rich models of the environment at a relative low cost in comparison with previous alternatives. Our SLAM approach is based on a hierarchical structure of keyframes with a low level layer of metric information and several topological layers intended for large scale SLAM and navigation. This SLAM solution, which makes use of the metric-topological representation mentioned above, works at video frame rate obtaining highly consistent maps. Future research is expected on metric-topological-semantic mapping from the new sensor and the SLAM system presented here. Finally, an extrinsic calibration technique is proposed to obtain the relative poses of a combination of 3D range sensors, like those employed in the omnidirectional RGB-D device mentioned above. The calibration is computed from the observation of planar surfaces of a structured environment in a fast, easy and robust way, presenting qualitative and quantitative advantages with respect to previous approaches. This technique is extended to calibrate any combination of range sensors, including 2D and 3D range sensors, in any configuration. The calibration of such sets of sensors is interesting not only for mobile robots, but also for autonomous cars
Learning to Represent Handwritten Shapes and Words for Matching and Recognition
Full text linkWriting is one of the most important forms of communication and for centuries, handwriting had been the most reliable way to preserve knowledge. However, despite the recent development of printing houses and electronic devices, handwriting is still broadly used for taking notes, doing annotations, or sketching ideas. In order to be easily accessed, there is a huge amount of handwritten documents, some of them with uncountable cultural value, that have been recently digitized. This has made necessary the development of methods able to extract information from these document images.Transferring the ability of understanding handwritten text or recognizing handwritten shapes to computers has been the goal of many researches due to its huge importance for many different fields. However, designing good representations to deal with handwritten shapes, e.g. symbols or words, is a very challenging problem due to the large variability of these kinds of shapes. One of the consequences of working with handwritten shapes is that we need representations to be, i.e., able to adapt to large intra-class variability. We need representations to be discriminative, i.e., able to learn what are the differences between classes. And, we need representations to be efficient, i.e., able to be rapidly computed and compared. Unfortunately, current techniques of handwritten shape representation for matching and recognition do not fulfill some or all of these requirements.Through this thesis we focus on the problem of learning to represent handwritten shapes aimed at retrieval and recognition tasks. Concretely, on the first part of the thesis, we focus on the general problem of representing any kind of handwritten shape. We first present a novel shape descriptor based on a deformable grid that deals with large deformations by adapting to the shape and where the cells of the grid can be used to extract different features. Then, we propose to use this descriptor to learn statistical models, based on the Active Appearance Model, that jointly learns the variability in structure and texture of a given class. Then, on the second part, we focus on a concrete application, the problem of representing handwritten words, for the tasks of word spotting, where the goal is to find all instances of a query word in a dataset of images, and recognition. First, we address the segmentation-free problem and propose an unsupervised, sliding-window-based approach that achieves state-of-the-art results in two public datasets.Second, we address the more challenging multi-writer problem, where the variability in words exponentially increases. We describe an approach in which both word images and text strings are embedded in a common vectorial subspace, and where those that represent the same word are close together. This is achieved by a combination of label embedding and attributes learning, and a common subspace regression. This leads to a low-dimensional, unified representation of word images and strings, resulting in a method that allows one to perform either image and text searches, as well as image transcription, in a unified framework. We evaluate our methods on different public datasets of both handwritten documents and natural images showing results comparable or better than the state-of-the-art on spotting and recognition tasks
Personal Identification Using Ears Based on Statistical Features
Full text linkBiometrics is an automated method of recognizing a person based on a physiological (e.g. face, iris, or retina) or behavioral (e.g. gait, signature, or dynamic keystrokes) characteristics. Ear recognition is one of the physiological biometrics\u27 types that have been interested in the recent years. Ear recognition, achieves good accuracy and has many advantages such as it doesn\u27t affected by expressions, health, and more stable than many other biometrics. However, it has many challenges such as the pose of the face, lighting variation, occlusion with hair or clothes. In this research, four proposed models are used to identify people using ear images. The first model used single feature extraction method based on single classifier. While, the second model used single feature extraction method based on multi-classifiers. The third model used feature combination techniques (parallel or serial) based on single classifier. Finally, in the fourth model multi-features and multi-classifiers are used. In this research, there are four methods that are used to extract the features, namely, \textit{Principal Component Analysis} (PCA), \textit{Linear Discriminant Analysis} (LDA), \textit{Independent Component Analysis} (ICA), and \textit{Discrete Cousin Transform} (DCT). Neural networks, decision tree, and minimum distance classifiers are used to classify the unknown samples. The occlusion problem with hair or scarves is one of the big challenges of the ear recognition systems. In this research, segmentation technique is proposed to neglect the occluded part and solve the occlusion problem. The idea of the segmentation technique is based on dividing the ear images into different parts. The occluded part/s is neglected and the rest of the parts are used to identify people based on features fusion and classifiers fusion. The segmentation technique consists of two main types, namely, uniform or non-uniform segmentation techniques. In this research, the uniform segmentation technique is used for many experiments (horizontal, vertical, and grid). All the four proposed models are applied to all ear segments to investigate the power of each model and to achieve a high accuracy. In this research, ear database images is used. The ear dataset consists of 102 grayscale images (6 images for each of 17 subjects) in PGM format [1]. The proposed models are achieved good identification rates using ear images. In the first model, the best accuracy achieved using LDA and neural network classifier. The results of the first model ranged from 64.12\% to 100\%. In the second model, many classifiers are fused to increase the recognition rate. In this method, two methods are used, namely, Borda count and majority voting. The results of this model ranged from 94.12\% to 96.08\%. The third model, the features using two different methods, namely serial and parallel are combined. The results of this model prove that the serial combination is more powerful than parallel combination. Finally, in the fourth model, two features and two classifiers are fused to get one decision. The accuracy of this model is approximately the same of the third model, and it does not achieve good results because there is a diversity between different classifiers. Moreover, the proposed segmentation model achieved good results when some parts of the ear images are occluded
Contextual Word Spotting in Historical Handwritten Documents
Full text linkThere are countless collections of historical documents in archives and libraries that contain plenty of valuable information for historians and researchers. The extraction of this information has become a central task among the Document Analysis researches and practitioners. There is an increasing interest to digital preserve and provide access to these kind of documents. But only the digitalization is not enough for the researchers. The extraction and/or indexation of information of this documents has had an increased interest among researchers. In many cases, and in particular in historical manuscripts, the full transcription of these documents is extremely difficult due the inherent deficiencies: poor physical preservation, different writing styles, obsolete languages, etc.Word spotting has become a popular an efficient alternative to full transcription. It inherently involves a high level of degradation in the images. The search of words is holistically formulated as a visual search of a given query shape in a larger image, instead of recognising the input text and searching the query word with an ascii string comparison. But the performance of classical word spotting approaches depend on the degradation level of the images being unacceptable in many cases . In this thesis we have proposed a novel paradigm called contextual word spotting method that uses the contextual/semantic information to achieve acceptable results whereas classical word spotting does not reach.The contextual word spotting framework proposed in this thesis is a segmentation-based word spotting approach, so an efficient word segmentation is needed. Historical handwritten documents present some common difficulties that can increase the difficulties the extraction of the words. We have proposed a line segmentation approach that formulates the problem as finding the central part path in the area between two consecutive lines. This is solved as a graph traversal problem. A path finding algorithm is used to find the optimal path in a graph, previously computed, between the text lines. Once the text lines are extracted, words are localized inside the text lines using a word segmentation technique from the state of the art.Classical word spotting approaches can be improved using the contextual information of the documents. We have introduced a new framework, oriented to handwritten documents that present a highly structure, to extract information making use of context. The framework is an efficient tool for semi-automatic transcription that uses the contextual information to achieve better results than classical word spotting approaches. The contextual information is automatically discovered by recognizing repetitive structures and categorizing all the words according to semantic classes. The most frequent words in each semantic cluster are extracted and the same text is used to transcribe all them.The experimental results achieved in this thesis outperform classical word spotting approaches demonstrating the suitability of the proposed ensemble architecture for spotting words in historical handwritten documents using contextual information
Multimodal Assessment of Shopping Behavior
Full text linkAutomatic understanding and recognition of human shopping behavior has many potential applications, attracting an increasing interest in the marketing domain. A first behavior cue regards the human movement patterns, then for obtaining a better overview of what is happening inside an environment, context information is used. More information regarding behavior can be extracted, by analyzing the interaction patterns with objects in the environment. Finally, facial expressions, which can be used to assess a person\u27s reaction to an object or in our case study to a product are employed as another informative behavior cue. Each intermediary analysis stream (trajectory analysis, action recognition, ROI detection module, and facial expression analysis), provides an input to the reasoning model, which based on the observables formulates a hypothesis regarding the most likely behavioral model. We integrated the different types of information on the semantic level, by implementing a multi-level framework. Finally, we evaluated this system in the ShopLab, in a real supermarket, and the product appreciation in a laboratory setting. The results show the feasibility of the approach in the recognition of trajectories (93%), shopping actions (91.6%), action units (93%), facial expressions (84%), and the most important behavioral types (87%)
Generalized Stacked Sequential Learning
Full text linkIn many supervised learning problems, it is assumed that data is independent and identically distributed. This assumption does not hold true in many real cases, where a neighboring pair of examples and their labels exhibit some kind of relationship. Sequential learning algorithms take benefit of these relationships in order to improve generalization. In the literature, there are different approaches that try to capture and exploit this correlation by means of different methodologies. In this thesis we focus on meta-learning strategies and, in particular, the stacked sequential learning (SSL) framework.The main contribution of this thesis is to generalize the SSL highlighting the key role of how to model theneighborhood interactions. We propose an effective and efficient way of capturing and exploiting sequentialcorrelations that take into account long-range interactions. We tested our method on several tasks: text lineclassification, image pixel classification, multi-class classification problems and human pose segmentation.Results on these tasks clearly show that our approach outperforms the standard stacked sequential learning as well as off-the-shelf graphical models such conditional random fields