Catálogo de publicaciones - libros

We address the problem of nonrigid Structure from Motion (SfM). Several methods have been published recently which try to solve the task of tracking, segmenting, or reconstructing nonrigid 3D objects in motion. Most of these papers focus on deformable objects. We deal with the segmentation of articulated objects, that is, nonrigid objects composed of several moving rigid objects. We consider two moving objects and assume that the rigid SfM problem has been solved for each of them separately. We propose a method which helps to decide whether an object is rotating around an axis defined by another moving object. The theories of the proposed method is discussed in detail. Experimental results for synthetic and real data are presented.

This paper presents a model for multiple cameras fusion, which is based on Dezert-Smarandache theory of evidence. We have developed a fusion model which works at the decision fusion level to track objects on a ground plane using geographically distributed cameras. As we are fusing at decision level, track is done based on predefined zones. We present early results of our model tested on CGI animated simulations, applying a perspective-based basic belief assignment function. Our experiments suggest that the proposed technique yields a good improvement in tracking accuracy when spatial regions are used to track.

The reconstruction and mapping of real scenes is a crucial element in several fields such as robot navigation. Stereo vision can be a powerful solution. However the perspective effect arises, as well as other problems, when the reconstruction is tackled using depth maps obtained from stereo images. A new approach is proposed to avoid the perspective effect, based on a geometrical rectification using the vanishing point of the image. It also uses sub-pixel precision to solve the lack of information for distant objects. Finally, the method is applied to map a whole scene, introducing a cubic filter.

In the paper spatial component estimation for Track–Before–Detect (TBD) based motion capture systems is presented. Using Likelihood Ratio TBD algorithm it is possible to track markers at low Signal–to-Noise Ratio level that is a typical case in motion capture system. Three kinds of TBD systems are analyzed and compared: full frame processing, single camera optimized and multiple cameras optimized. In the article separate TBD processing for every camera is assumed.

We present a fully automatic real-time algorithm for robust and accurate left ventricular segmentation in three-dimensional (3D) cardiac ultrasound. Segmentation is performed in a sequential state estimation fashion using an extended Kalman filter to recursively predict and update the parameters of a 3D Active Shape Model (ASM) in real-time. The ASM was trained by tracing the left ventricle in 31 patients, and provided a compact and physiological realistic shape space. The feasibility of the proposed algorithm was evaluated in 21 patients, and compared to manually verified segmentations from a custom-made semi-automatic segmentation algorithm. Successful segmentation was achieved in all cases. The limits of agreement (mean±1.96SD) for the point-to-surface distance were 2.2±1.1mm. For volumes, the correlation coefficient was 0.95 and the limits of agreement were 3.4±20 ml. Real-time segmentation of 25 frames per second was achieved with a CPU load of 22%.

Early detection of glaucoma is essential for preventing one of the most common causes of blindness. Our research is focused on a novel automated classification system based on image features from fundus photographs which does not depend on structure segmentation or prior expert knowledge. Our new data driven approach that needs no manual assistance achieves an accuracy of detecting glaucomatous retina fundus images compareable to human experts. In this paper, we study image preprocessing methods to provide better input for more reliable automated glaucoma detection. We reduce disease independent variations without removing information that discriminates between images of healthy and glaucomatous eyes. In particular, nonuniform illumination is corrected, blood vessels are inpainted and the region of interest is normalized before feature extraction and subsequent classification. The effect of these steps was evaluated using principal component analysis for dimension reduction and support vector machine as classifier.

The MET protein controls growth, invasion, and metastasis in cancer cells and is thereby of interest to study, for example from a structural point of view. For individual particle imaging by Cryo-Electron Tomography of the MET protein, or other proteins, dedicated image analysis methods are required to extract information in a robust way as the images have low contrast and resolution (with respect to the size of the imaged structure). We present a method to identify the two parts of the MET protein, -propeller and stalk, using a fuzzy framework. Furthermore, we describe how a representation of the MET stalk, denoted , can be identified based on the use of non-uniform B-splines. The stalk curve is used to extract relevant geometrical information about the stalk, e.g., to facilitate curvature and length measurements.

Navigation through large microscopical images is a potential benefit for histology or pathology teaching, quality of diagnosis in pathology or communication between pathologists in some telemedicine applications. However, the size of this kind of images is prohibitive for navigation with conventional techniques. This paper presents a system which constructs large microscopical images and describes optimal strategies for navigation using the JPEG2000 (J2K) standard. Given a set of images automatically acquired from a microscope, they are registered to construct a mega-image. Using the J2K standard the image is compressed and stored. Finally, we developed a novel technique for browsing the image from the compressed bit stream, which allows to efficiently obtain regions of interest with different qualities and resolutions, i.e. in microscopical terms, with different enlargement settings.

An automatic classification system into either malignant or benign microcalcification from mammograms is a helpful tool in breast cancer diagnosis. From a set of extracted features, a classifying method using neural networks can provide a probability estimation that can help the radiologist in his diagnosis. With this objective in mind, this paper proposes a feature selection algorithm from a massive number of features based on a statistical distance method in conjunction with a genetic algorithm (GA). The use of a statistical distance as optimality criterion was improved with genetic algorithms for selecting an appropriate subset of features, thus making this algorithm capable of performing feature selection from a massive set of initial features. Additionally, it provides a criterion to select an appropriate number of features to be employed. Experimental work was performed using Generalized Softmax Perceptrons (GSP), trained with a Strict Sense Bayesian cost function for direct probability estimation, as microcalcification classifiers. A Posterior Probability Model Selection (PPMS) algorithm was employed to determine the network complexity. Results showed that this algorithm converges into a subset of features which has a good classification rate and Area Under Curve (AUC) of the Receiver Operating Curve (ROC).

Segmentation of femurs in Anterior-Posterior x-ray images is very important for fracture detection, computer-aided surgery and surgical planning. Existing methods do not perform well in segmenting bones in x-ray images due to the presence of large amount of spurious edges. This paper presents an atlas-based approach for automatic segmentation of femurs in x-ray images. A robust global alignment method based on consistent sets of edge segments registers the whole atlas to the image under joint constraints. After global alignment, the femur models undergo local refinement to extract detailed contours of the femurs. Test results show that the proposed algorithm is robust and accurate in segmenting the femur contours of different patients.