Catálogo de publicaciones - libros

In this paper, we address the problem of ink parsing, which tries to identify distinct symbols from a stream of pen strokes. An important task of this process is the segmentation of the users’ pen strokes into salient fragments based on geometric features. This process allows users to create a sketch symbol varying the number of pen strokes, obtaining a more natural drawing environment. The proposed sketch recognition technique is an extension of LR parsing techniques, and includes ink segmentation and context disambiguation. During the parsing process, the strokes are incrementally segmented by using a dynamic programming algorithm. The segmentation process is based on templates specified in the productions of the grammar specification from which the parser is automatically constructed.

An approach for segmentation of ultrasound images using features extracted by orthogonal wavelet transforms that can be used in an interactive system is proposed. These features are the training data for the K-means clustering algorithm and the Bayes classifier. The result of classification is improved by using neighbourhood information.

Caricature is emphasizing the distinctive features of a particular face. Exaggerating the Difference from the Mean (EDFM) is widely accepted among caricaturists to be the driving factor behind caricature generation. However the caricatures created by different artists have different drawing style. No attempt has been taken in the past to identify these distinct drawing styles. Yet the proper identification of the drawing style of an artist will allow the accurate modelling of a personalised exaggeration process, leading to fully automatic caricature generation with increased accuracy. In this paper we provide experimental results and detailed analysis to prove that a Cascade Correlation Neural Network (CCNN) can be used for capturing the drawing style of an artist and thereby used in realistic automatic caricature generation. This work is the first attempt to use neural networks in this application area and have the potential to revolutionize existing automatic caricature generation technologies.

Most approaches to text classification rely on some measure of (dis)similarity between sequences of symbols. Information theoretic measures have the advantage of making very few assumptions on the models which are considered to have generated the sequences, and have been the focus of recent interest. This paper addresses the use of the Ziv-Merhav method (ZMM) for the estimation of relative entropy (or Kullback-Leibler divergence) from sequences of symbols as a tool for text classification. We describe an implementation of the ZMM based on a modified version of the Lempel-Ziv algorithm (LZ77). Assessing the accuracy of the ZMM on synthetic Markov sequences shows that it yields good estimates of the Kullback-Leibler divergence. Finally, we apply the method in a text classification problem (more specifically, authorship attribution) outperforming a previously proposed (also information theoretic) method.

Finite-state models are used to implement a handwritten text recognition and classification system for a real application entailing casual, spontaneous writing with large vocabulary. Handwritten short phrases which involve a wide variety of writing styles and contain many non-textual artifacts, are to be classified into a small number of predefined classes. To this end, two different types of statistical framework for phrase recognition-classification are considered, based on finite-state models. HMMs are used for text recognition process. Depending to the considered architecture, N -grams are used for performing text recognition and then text classification (serial approach) or for performing both simultaneously (integrated approach). The multinomial text classifier is also employed in the classification phase of the serial approach. Experimental results are reported which, given the extreme difficulty of the task, are encouraging.

In this paper, a new approach to binarize grey-level document images is proposed. The method combines a global and a local approaches. First, we provide the edges of the image, and next, from the edges we make a quadtree decomposition of the image. On each area of the image, a local threshold is computed and applied to all the pixels belonging to the region under consideration.

Usually, in traditional text categorization systems based on Vector Space Model, there is no context information in a feature vector, which limited the performance of the system. To make use of more information, it is natural to select bi-gram feature in addition to unigram feature. However, the longer the feature is, the more important the feature selection algorithm is to get good balance in feature space This paper proposed two feature extraction methods which can get better feature balance for document categorization. Experiments show that our extended bi-gram feature improved system performance greatly.

In this paper we tackle the specific problem of old documents recovery. Spots, print through, underlines and others ageing features are undesirable not only because they harm the visual appearance of the document, but also because they affect future Optical Character Recognition (OCR). This paper proposes a new method integrating fuzzy clustering of color properties of original images and mathematical morphology. We will show that this technique leads to higher quality of the recovered images and, at the same time, it delivers cleaned binary text for OCR applications. The proposed method was applied to books of XIX Century, which were cleaned in a very effective way.

In pattern matching based Protein-Protein Interaction Extraction systems, patterns generated manually or automatically exist erroneous and redundancy, which greatly affect the system’s performance. In this paper, a MDL-based pattern optimizing algorithm is proposed to filter out the bad patterns and redundancy. Experiments show that our algorithm is effective in improving the system’s performance while greatly cutting down the number of patterns. It also has excellent generalizability which is important in implementing practical systems.

Clustering is a technique extensively employed for the analysis, classification and annotation of DNA microarrays. In particular clustering based upon the classical combinatorial curvature is widely applied. We introduce a new clustering method for vertex-weighted networks, method which is based upon a generalization of the combinatorial curvature. The new measure is of a geometric nature and represents the metric curvature of the network, perceived as a finite metric space. The metric in question is natural one, being induced by the weights. We apply our method to publicly available yeast and human lymphoma data. We believe this method provides a much more delicate, graduate method of clustering then the other methods which do not undertake to ascertain all the relevant data. We compare our results with other works. Our implementation is based upon Trixy (as available at http://tagc.univ-mrs.fr/bioinformatics/trixy.html ), with some appropriate modifications to befit the new method.