Catálogo de publicaciones - libros

Many of the watermarking schemes that claim resilience to geometrical distortions embed information into invariant or semi-invariant domains. However, the discretisation process required in such domains might lead to low correlation responses during watermarking detection. In this document, a new strategy is proposed to provide resilience to strong Rotation, Scaling and Translation (RST) distortions. The proposed detection process is based on a Genetic Algorithm (GA) that maximises the correlation coefficient between the originally embedded watermark and the input image. Comparisons between a previous scheme, based on Log-Polar Mapping (LPM), and the present approach are reported. Results show that even a simple insertion process provides more robustness, as well as a lower image degradation.

In this research we propose to use phoneme spotting to improve the results in the generation of a cryptographic key. Phoneme spotting selects the phonemes with highest accuracy in the user classification task. The key bits are constructed by using the Automatic Speech Recognition and Support Vector Machines. Firstly, a speech recogniser detects the phoneme limits in each speech utterance. Afterwards, the support vector machine performs a user classification and generates a key. By selecting the highest accuracy phonemes for a a set of 10, 20, 30 and 50 speakers randomly chosen from the YOHO database, it is possible to generate reliable cryptographic keys.

In this work is proposed a new approach based on Evolving Fuzzy Neural Networks (EFuNNs) to on-line evaluation of training in virtual reality worlds. EFuNNs are dynamic connectionist feed forward networks with five layers of neurons and they are adaptive rule-based systems. Results of the technique application are provided and compared with another evaluation system based on a backpropagation trained multilayer perceptron neural network.

The goal of this work is to automatically determine the level of tool insert wear based on images acquired using a vision system. Experimental wear was carried out by machining AISI SAE 1045 and 4140 steel bars in a precision CNC lathe and using Sandvik inserts of tungsten carbide. A Pulnix PE2015 B/W with an optic composed by an industrial zoom 70 XL to 1.5X and a diffuse lighting system was used for acquisition. After images were pre-processed and wear area segmented, several patterns of the wear area were obtained using a set of descriptors based on statistical moments. Two sets of experiments were carried out, the first one considering two classes, low wear level and high wear level, respectively; the second one considering three classes. Performance of three classifiers was evaluated: Lp, k-nearest neighbours and neural networks. Zernike and Legendre descriptors show the lowest error rates using a MLP neuronal network for classifying.

In this paper, we propose a robust surface registration using a Gaussian-weighted distance map for PET-CT brain fusion. Our method is composed of three steps. First, we segment the head using the inverse region growing and remove the non-head regions segmented with the head using the region growing-based labeling in PET and CT images, respectively. The feature points of the head are then extracted using sharpening filter. Second, a Gaussian-weighted distance map is generated from the feature points of CT images to lead our similarity measure to robust convergence on the optimal location. Third, weighted cross-correlation measures the similarities between the feature points extracted from PET images and the Gaussian-weighted distance map of CT images. In our experiments, we use software phantom and clinical datasets for evaluating our method with the aspect of visual inspection, accuracy, robustness, and computation time. Experimental results show that our method is more accurate and robust than the conventional ones.

Locating the minimum number of sensors able to see at the same time the entire surface of an object is an important practical problem. Most work presented in this area is restricted to 2D objects. In this paper we present an optimal 3D sensor location algorithms that can locate sensors into a polyhedral environment that are able to see the features of the objects in their entirety. Limitations due to real sensors can be easily taken into account. The algorithm has been implemented, and examples are also given.

This paper generalizes the technique described in [1] to gray-scale image processing applications. This method chooses a subset of variables (i.e. pixels seen through a window) that maximizes the information observed in a set of training data by mean conditional entropy minimization. The task is formalized as a combinatorial optimization problem, where the search space is the powerset of the candidate variables and the measure to be minimized is the mean entropy of the estimated conditional probabilities. As a full exploration of the search space requires an enormous computational effort, some heuristics of the feature selection literature are applied. The introduced approach is mathematically sound and experimental results with texture recognition application show that it is also adequate to treat problems with gray-scale images.

In this paper we consider some questions related to the orientation of shapes when the standard method does not work. A typical situation is when a shapes under consideration has more than two axes of symmetry or if the shape is -fold rotationally symmetric, when > 2. Those situations are well studied in literature. Here, we give a very simple proof of the main result from [11] and slightly adapt their definition of principal axes for rotationally symmetric shapes. We show some desirable properties that hold if the orientation of such shapes is computed in such a modified way.

The analysis of spatial relations among objects in an image is a important vision problem that involves both shape analysis and structural pattern recognition. In this paper, we propose a new approach to characterize the spatial relation , an important feature of spatial configuration in space that has been overlooked in the literature up to now. We propose a mathematical definition of the degree to which an object is along an object , based on the region and and a degree of elongatedness of this region. In order to better fit the perceptual meaning of the relation, distance information is included as well. Experimental results obtained using synthetic shapes and brain structures in medical imaging corroborate the proposed model and the derived measures, thus showing their adequation with the common sense.

This paper presents a computational model for pattern analysis and classification using symmetry group theory. The model was designed to be part of an integrated management system for pattern design cataloguing and retrieval in the textile and tile industries. While another reference model [6], uses intensive image processing operations, our model is oriented to the use of graphic entities. The model starts by detecting the objects present in the initial digitized image. These objects are then transformed into Bezier curves and grouped to form motifs. The objects and motifs are compared and their symmetries are computed. Motif repetition in the pattern provides the fundamental parallelogram, the deflexion axes and rotation centres that allow us to classify the pattern according its plane symmetry group. This paper summarizes the results obtained from processing 22 pattern designs from Islamic mosaics in the Alcazar of Seville.