Catálogo de publicaciones - libros

Registration of CT and PET thoracic images has to cope with deformations of the lungs during breathing. Possible tumors in the lungs usually do not follow the same deformations, and this should be taken into account in the registration procedure. We show in this paper how to introduce tumor-based constraints into a non-linear registration of thoracic CT and PET images. Tumors are segmented by means of a semi-automatic procedure and they are used to guarantee relevant deformations near the pathology. Results on synthetic and real data demonstrate a significant improvement of the combination of anatomical and functional images for diagnosis and for oncology applications.

This paper presents a new approach to the question of surface grading based on and well known classifiers. These descriptors come from global image statistics computed in perceptually uniform colour spaces (CIE Lab or CIE Luv). The method has been extracted and validated using a statistical procedure based on and . The method is not a new theoretical contribution, but we have found and demonstrate that a simple set of global statistics softly describing colour and texture properties, together with well-known classifiers, are powerful enough to meet stringent factory requirements for real-time and performance. These requirements are on-line inspection capability and 95% surface grading accuracy. The approach is also compared with two other methods in the surface grading literature; colour histograms [1] and centile-LBP [8]. This paper is an extension and in-depth development of ideas reported in a previous work [11].

An effective algorithm for automatic removal impulse noise from highly corrupted monochromatic images is proposed. The method consists of two steps. Outliers are first detected using local spatial relationships between image pixels. Then the detected noise pixels are replaced with the output of a rank-order filter over a local spatially connected area excluding the outliers, while noise-free pixels are left unaltered. Simulation results in test images show a superior performance of the proposed filtering algorithm comparing with conventional filters. The comparisons are made using mean square error, mean absolute error, and subjective human visual error criterion.

We have developed a -continuous cubic A-spline, suitable for smoothing polygonal chains used in 2D shape representation. The proposed A-spline scheme interpolates an ordered set of data points in the plane, as well as the direction and sense of tangent vectors associated to these points. We explicitly characterize curve families which are used to construct the A-spline sections, whose members have the required interpolating properties and possess a minimal number of inflection points. The A-spline considered here has many attractive features: it is very easy to construct, it provides us with convenient geometric control handles to locally modify the shape of the curve and the error of approximation is controllable. Furthermore, it can be rapidly displayed, even though its sections are implicitly defined algebraic curves.

Mathematics Subject Classification: 65D07(splines), 65D05 (interpolation), 65D17 (Computer Aided Design).

In this paper, a robust statistical model-based brain MRI image segmentation method is presented. The MRI images are modeled by Gaussian mixture model. This method, based on the statistical model, approximately finds the maximum a posteriori estimation of the segmentation and estimates the model parameters from the image data. The proposed strategy for segmentation is based on the EM and FCM algorithm. The prior model parameters are estimated via EM algorithm. Then, in order to obtain a good segmentation and speed up the convergence rate, initial estimates of the parameters were done by FCM algorithm. The proposed image segmentation methods have been tested using phantom simulated MRI data. The experimental results show the proposed method is effective and robust.

A new version of the algorithm, called , is presented. The main difference between them is the capability of the new one to extend the training set during the inference process. The effect of this new feature is specially notorious in the inference of languages generated from regular expressions and Non-deterministic Finite Automata (NFA). A first experimental comparison is done between and , other algorithm that behaves well with the same sort of training data.

A computational approach to the perception of illusory contours is introduced. The approach is based on the technique and applied to several real and synthetic images. Special interest is given to the design of the communication pattern for spatial contour integration, called voting field.

In this article, the clustering problem under the criterion of minimum sum of squares clustering is considered. It is known that this problem is a nonconvex program which possesses many locally optimal values, resulting that its solution often falls into these traps. To explore the proper result, a novel clustering technique based on improved noising method called INMC is developed, in which one-step DHB algorithm as the local improvement operation is integrated into the algorithm framework to fine-tune the clustering solution obtained in the process of iterations. Moreover, a new method for creating the neighboring solution of the noising method called mergence and partition operation is designed and analyzed in detail. Compared with two noising method based clustering algorithms recently reported, the proposed algorithm greatly improves the performance without the increase of the time complexity, which is extensively demonstrated for experimental data sets.

This paper presents the results obtained in a real experiment for object recognition in a sequence of images captured by a mobile robot in an indoor environment. The purpose is that the robot learns to identify and locate objects of interest in its environment from samples of different views of the objects taken from video sequences. In this work, objects are simply represented as an unstructured set of spots (image regions) for each frame, which are obtained from the result of an image segmentation algorithm applied on the whole sequence. Each spot is semi-automatically assigned to a class (one of the objects or the background) and different features (color, size and invariant moments) are computed for it. These labeled data are given to a feed-forward neural network which is trained to classify the spots. The results obtained with all the features, several feature subsets and a backward selection method show the feasibility of the approach and point to color as the fundamental feature for discriminative ability.

Polarization of light caused by reflection from dielectric surfaces has been widely studied in computer vision. This paper presents an analysis of the accuracy of a technique that has been developed to acquire surface orientation from the polarization state of diffusely reflected light. This method employs a digital camera and a rotating linear polarizer. The paper also explores the possibility of linking polarization vision with shading information by means of a computationally efficient BRDF estimation algorithm.