Catálogo de publicaciones - libros

In medical diagnosis symptoms often differ in nature varying form linguistic information trough numerical values to crisp statements. Hence, unification of their interpretation during diagnosis support is difficult. The authors propose the diagnosis support using an extension of the Dempster-Shafer theory for fuzzy focal elements. Performance of the method depends on the membership function shapes. The paper provides indications for imprecise symptom representation and its membership function determination. The problems are discussed for thyroid gland diseases. Conclusions helpful in diagnosis support are formulated.

Search Engines of today have evolved from generic ones that find matches based on keywords, to those that provide personalized search based on concepts specified explicitly by the users. In this paper we present an approach to build dynamic user model for personalized search based on the user’s browsing history using Open Directory Project Concept hierarchy that not only learns user’s interest implicitly, but also tracks the temporal evolution and digression of their interests and modifies their profile accordingly.

To discuss the relationship between and , a formal definition of the as well as a uniform approach to the all methods are proposed first, and then a problem is introduced and formalized. The equivalence relationship between and , thoroughly considered and illustrated by examples, proves to allow incorporation of the methods and techniques specific for the approach to the and vice versa.

Linear discrimination analysis (LDA) technique is an important and well-developed area of image recognition and to date many linear discrimination methods have been put forward. Despite these efforts, there persist in the traditional LDA some weaknesses. In this paper, we propose a new LDA-based method that can overcome the drawback existed in the traditional LDA method. It redefines the between-class scatter by adding a weight function according to the between-class distance, which helps to separate the classes as much as possible. At the same time, in this method, we firstly remove the null space of total scatter matrix which has been proved to be the common null space of both between-class and within-class scatter matrix, and useless for discrimination. Then in the lower-dimensional projected space, the null space of the resulting within-class scatter matrix is calculated. This lower-dimensional null space, combined with the previous projection, represents a subspace of the whole null space of within-class scatter matrix, and is really useful for discrimination. The optimal discriminant vectors of LDA are derived from it. Experiment results show our method achieves better performance in comparison with the traditional LDA methods.

In this paper we consider some problems related to attribute dependencies in consensus determining. These problems concern the dependencies of attributes representing the content of conflicts, which cause that one may not treat the attributes independently in consensus determining. We show how to cope with situations when consensus values calculated for the attributes do not fulfill the dependency function: an error estimation method and an algorithm for calculating a consensus for dependent attribute are presented.

In this paper the problems of sequential pattern recognition are considered. As a statistical model of dependence, in the sequences of patterns, the firstorder Markov chain is assumed. Additionally, the assumption about independence between the attributes in the feature vector is made. The pattern recognition algorithms with such assumption are called in the literature “naive Bayes algorithm”. In this paper such approach is made to the pattern recognition algorithm for first-order Markov chain and some results of numerical investigation are presented. The main goal of these investigations was to verify if it is reasonable to make such assumption in the real recognition tasks.

This paper presents an approach based on evolutionary computations to the IFS inverse problem. Having a bitmap image we look for a set of functions that can reproduce a good approximation if a given image. A method using variable number of mappings is proposed. A number of different crossover operators is described and tested. Different parameters for fitness functions are also tested. The paper ends with some experimental results showing images we were able to generate with our method

The recently proposed algorithm, using the evidence accumulation principle, for finding lines (ridges) having shape which can be neither parameterized nor tabularized is described in detail. This fuzzy, multi-scale algorithm stores the evidence in the accumulator congruent with the image domain. The primary application was finding blood vessels in mammograms.

The scale and rotation invariance properties of a recently proposed algorithm, using the fuzzy evidence accumulation principle, for finding lines (ridges) of non-parametric shapes is analysed. The proposed modifications consist in scaling the accumulated value with the inverse of the line width and further fuzzifying the accumulation process — along the line width. Good invariance properties received are tested on artificial images and confirmed on real-life mammographic images.

A retrieval methodology which integrates color, texture and shape information is presented in this paper. Consequently, the overall image similarity is developed through the similarity based on all the feature components. Alternatively to known CBIR systems, we compute features only in the finite number of extracted ROIs. There are some other known methods of determining ROIs, but our method of extracting ROI based on points of interest detection and Gabor filtration, enables to use filter responses also to describe texture parameters. The described method was tested on a small post stamps database (130 stamps), for which we achieved comparable results as for system. Presented method is further developed in postal image analysis and retrieval system.