Catálogo de publicaciones - libros

Given an appropriate imaging resolution, a common Magnetic Resonance Imaging (MRI)model assumes that the object under study is composed of piecewise constant materials, so that MRI produces piecewise constant images. The intensity inhomogeneity (IIH), due to the spatial inhomogeneity in the excitatory Radio Frequency (RF) signal and other effects, is modeled by a multiplicative inhomogeneity field. We propose and test two estimation rules of the IIH field, inspired in the Self Organizing Map (SOM), derived from well defined energy functions.

The correct determination of the position of incident photons is a crucial issue in PET imaging.In this paper we study the use of Neural Networks (NNs) for position estimation of photons impinging on gamma-ray detector modules for PET cameras based on continuous scintillators and Multi-Anode Photomultiplier Tubes (MA-PMTs). We have performed a thorough analysis of the NN architecture and training procedures, using realistic simulated inputs, in order to achieve the best results in terms of spatial resolution and bias correction. The results confirm that NNs can partially model and correct the non-uniform detector response using only the position-weighted signals from a simple 2D Discretized Positioning Circuit (DPC). Linearity degradation for oblique incidence is also investigated. Finally, the NN can be implemented in hardware for parallel real time corrected Line-of-Response (LOR) estimation.

In this article we present the results of a study carried out using the popular fastica algorithm applied to the detection of filters in natural images in gray-scale, contaminated with gaussian noise. The detection of filters has been accomplished by using the statistical distribution measures kurtosis and skewness.

In this paper, we propose a novel approach for facial expression analysis and recognition. The main contributions of the paper are as follows. First, we propose an efficient facial expression recognition scheme based on the detection of keyframes in videos where the recognition is performed using a temporal classifier. Second, we use the proposed method for extending the human-machine interaction functionality of the AIBO robot. More precisely, the robot is displaying an emotional state in response to the recognized user’s facial expression. Experiments using unseen videos demonstrated the effectiveness of the developed method.

Face recognition represents one of the most interesting modalities of biometric. Due to his low intrusiveness and to the constant decrease in image acquisition cost, it’s particularly suitable for a wide number of real time applications. In this paper we propose a very fast image pre-processing by the introduction of a linearly shaded elliptical mask centered over the faces. Used in association with DCT, for features extraction, and MPL and RBF Neural Networks, for classification, it allows an improvement of system performances without modifying the global computation weight and also a learning time reduction for MLP neural networks.

Address-Event-Representation (AER) is a communication protocol for transferring spikes between bio-inspired chips. Such systems may consist of a hierarchical structure with several chips that transmit spikes among them in real time, while performing some processing. There exist several AER tools to help in developing and testing AER based systems. These tools require the use of a computer to allow the processing of the event information, reaching very high bandwidth at the AER communication level. We propose to use an embedded platform based on multi-task operating system to allow both, the AER communication and the AER processing without a laptop or a computer. We have connected and programmed a Gumstix computer to process Address-Event information and measured the performance referred to the previous AER tools solutions. In this paper, we present and study the performance of a new philosophy of a frame-grabber AER tool based on a multi-task environment, composed by the Intel XScale processor governed by an embedded GNU/Linux system.

Automatic road sign recognition systems require a great computational cost since the number of different signs in each country is quite large. In many real-world applications only a reduced subset of traffic signs is considered in the recognition stage to verify the success of a classification algorithm. This paper proposes a optimization in the traffic sign identification task working in the spatial domain. This purpose is overcome through dimension reduction methods, such as 2DPCA and (2D)PCA, to perform principal component analysis of training and test image vectors. The applications of these advances, using SVMs as classification technique, are shown here over a considerable database.

Several complex problems have to be solved in order to build Advanced Driving Assistance Systems. Among them, an important problem is the detection and classification of traffic signs, which can appear at any position within a captured image. This paper describes a system that employs independent modules to classify several prohibition road signs. Combining the predictions made by the set of classifiers, a unique final classification is achieved. To reduce the computational complexity and to achieve a real-time system, a previous input feature selection is performed. Experimental evaluation confirms that using this feature selection allows a significant input data reduction without an important loss of output accuracy.

The aim of this work is to design a Traffic Sign Classification system that combines different image preprocessing techniques with Neural Networks. It must be robust against image problems like rotation, deterioration, vandalism, and so on. The preprocessings applied to the gray scale transformed image are: the median filter (MF), the histogram equalization (HE), and the vertical (VH) and horizontal (HH) histograms with fixed or variable (mean value or Otsu method) thresholding. The k-Nearest Neighbour (k-NN) classifier is used for comparison purposes. The best performance is obtained with the combination of preprocessings: MF, HE and VH and HH with a fixed threshold ( = 185), with a two hidden layer MultiLayer Perceptron (MLP), which achieves a probability of classification of 98,72% for nine different classes of blue traffic signs and noise. The performance is better than the classifier based on one hidden layer MLP in at least 1,28% and based on k-NN in at least 5,13%. If computational cost must be reduced, other preprocessings with a one hidden layer MLP are proposed, which performance is lower.

In this work, a two-dimensional (2-D) representation of the hourly solar radiation data is proposed. The model enables accurate forecasting using image prediction methods. One year solar radiation data that is acquired and collected between August 1, 2005 and July 30, 2006 in Iki Eylul campus of Anadolu University, and a 2-D representation is formed to construct an image data. The data is in raster scan form, so the rows and columns of the image matrix indicate days and hours, respectively. To test the forecasting efficiency of the model, first 1-D and 2-D optimal 3-tap linear filters are calculated and applied. Then, the forecasting is tested through three input one output feed−forward neural networks (NN). One year data is used for training, and 2 month(from August 1,2006 to September 30,2006) for testing. Optimal linear filters and NN models are compared in the sense of root mean square error (RMSE). It is observed that the 2-D model has advantages over the 1-D representation. Furthermore, the NN model accurately converges to forecasting errors smaller than the linear prediction filter results.