Abstract: Wireless endoscopy is a very recent and at the same time unique technique allowing to visualize and study the occurrence of con- tractions and to analyze the intestine motility. Feature extraction is es- sential for getting efficient patterns to detect contractions in wireless video endoscopy of small intestine. We propose a novel method based on anisotropic image filtering and efficient statistical classification of con- traction features. In particular, we apply the image gradient tensor for mining informative skeletons from the original image and a sequence of descriptors for capturing the characteristic pattern of contractions. Fea- tures extracted from the endoluminal images were evaluated in terms of their discriminatory ability in correct classifying images as either belong- ing to contractions or not. Classification was performed by means of a support vector machine classifier with a radial basis function kernel. Our classification rates gave sensitivity of the order of 90.84% and specificity of the order of 94.43% respectively. These preliminary results highlight the high efficiency of the selected descriptors and support the feasibility of the proposed method in assisting the automatic detection and analysis of contractions.

Abstract: Small Bowel Motility Assessment by means of Wireless Capsule Video Endoscopy constitutes a novel clinical methodology in which a capsule with a micro-camera attached to it is swallowed by the patient, emitting a RF signal which is recorded as a video of its trip throughout the gut. In order to overcome the main drawbacks associated with this technique -mainly related to the large amount of visualization time required-, our efforts have been focused on the development of a machine learning system, built up in sequential stages, which provides the specialists with the useful part of the video, rejecting those parts not valid for analysis. We successfully used Self Organized Maps in a general semi-supervised framework with the aim of tackling the different learning stages of our system. The analysis of the diverse types of images and the automatic detection of intestinal contractions is performed under the perspective of intestinal motility assessment in a clinical environment.

Abstract: This article describes a novel approach to the modeling of human actions in 3D. The method we propose is based on a “bag of poses” model that represents human actions as histograms of key-pose occurrences over the course of a video sequence. Actions are first represented as 3D poses using a sequence of 36 direction cosines corresponding to the angles 12 joints form with the world coordinate frame in an articulated human body model. These pose representations are then projected to three-dimensional, action-specific principal eigenspaces which we refer to as aSpaces. We introduce a method for key-pose selection based on a local-motion energy optimization criterion and we show that this method is more stable and more resistant to noisy data than other key-poses selection criteria for action recognition.

Abstract: Poor convergence to concave shapes is a main limitation of snakes as a standard segmentation and shape modelling technique. The gradient of the external energy of the snake represents a force that pushes the snake into concave regions, as its internal energy increases when new inexion points are created. In spite of the improvement of the external energy by the gradient vector ow technique, highly non convex shapes can not be obtained, yet. In the present paper, we develop a new external energy based on the geometry of the curve to be modelled. By tracking back the deformation of a curve that evolves by minimum curvature ow, we construct a distance map that encapsulates the natural way of adapting to non convex shapes. The gradient of this map, which we call curvature vector ow (CVF), is capable of attracting a snake towards any contour, whatever its geometry. Our experiments show that, any initial snake condition converges to the curve to be modelled in optimal time.

Abstract: Graph embeddings in vector spaces aim at assigning a pattern vector to every graph so that the problems of graph classification and clustering can be solved by using data processing algorithms originally developed for statistical feature vectors. An important requirement graph features should fulfil is that they reproduce as much as possible the properties among objects in the graph domain. In particular, it is usually desired that distances between pairs of graphs in the graph domain closely resemble those between their corresponding vectorial representations. In this work, we analyse relations between the edit distance in the graph domain and the L1 distance of the attribute statistics based embedding, for which good classification performance has been reported on various datasets. We show that there is actually a high correlation between the two kinds of distances provided that the corresponding parameter values that account for balancing the weight between node and edge based features are properly selected.

Abstract: This work develops a new architecture for multiple-target tracking in unconstrained dynamic scenes, which consists of a detection level which feeds a two-stage tracking system. A remarkable characteristic of the system is its ability to track several targets while they group and split, without using 3D information. Thus, special attention is given to the feature-selection and appearance-computation modules, and to those modules involved in tracking through groups. The system aims to work as a stand-alone application in complex and dynamic scenarios. No a-priori knowledge about either the scene or the targets, based on a previous training period, is used. Hence, the scenario is completely unknown beforehand. Successful tracking has been demonstrated in well-known databases of both indoor and outdoor scenarios. Accurate and robust localisations have been yielded during long-term target merging and occlusions.

Abstract: Traffic sign classification represents a classical application of multi-object recognition processing in uncontrolled adverse environments. Lack of visibility, illumination changes, and partial occlusions are just a few problems. In this paper, we introduce a novel system for multi-class classification of traffic signs based on error correcting output codes (ECOC). ECOC is based on an ensemble of binary classifiers that are trained on bi-partition of classes. We classify a wide set of traffic signs types using robust error correcting codings. Moreover, we introduce the novel β-correction decoding strategy that outperforms the state-of-the-art decoding techniques, classifying a high number of classes with great success.

Abstract: This paper deals with the topic of performance evaluation of symbol recognition & spotting systems. We propose here a new approach to the generation of synthetic graphics documents containing non-isolated symbols in a real context. This approach is based on the definition of a set of constraints that permit us to place the symbols on a pre-defined background according to the properties of a particular domain (architecture, electronics, engineering, etc.). In this way, we can obtain a large amount of images resembling real documents by simply defining the set of constraints and providing a few pre-defined backgrounds. As documents are synthetically generated, the groundtruth (the location and the label of every symbol) becomes automatically available. We have applied this approach to the generation of a large database of architectural drawings and electronic diagrams, which shows the flexibility of the system. Performance evaluation experiments of a symbol localization system show that our approach permits to generate documents with different features that are reflected in variation of localization results.