Informatica

The article presents a limited‐vocabulary speaker independent continuous Estonian speech recognition system based on hidden Markov models. The system is trained using an annotated Estonian speech database of 60 speakers, approximately 4 hours in duration. Words are modelled using clustered triphones with multiple Gaussian mixture components. The system is evaluated using a number recognition task and a simple medium‐vocabulary recognition task. The system performance is explored by employing acoustic models of increasing complexity. The number recognizer achieves an accuracy of 97%. The medium‐vocabulary system recognizes 82.9% words correctly if operating in real time. The correctness increases to 90.6% if real‐time requirement is discarded.

This paper presents a new in‐place pseudo linear radix sorting algorithm. The proposed algorithm, called MSL (Map Shuffle Loop) is an improvement over ARL (Maus, 2002). The ARL algorithm uses an in‐place permutation loop of linear complexity in terms of input size. MSL uses a faster permutation loop searching for the next element to permute group by group, instead of element by element. The algorithm and its runtime behavior are discussed in detail. The performance of MSL is compared with quicksort and the fastest variant of radix sorting algorithms, which is the Least Significant Digit (LSD) radix sorting algorithm (Sedgewick, 2003).

The paper describes a new method to segment ischemic stroke region on computed tomography (CT) images by utilizing joint features from mean, standard deviation, histogram, and gray level co‐occurrence matrix methods. Presented unsupervised segmentation technique shows ability to segment ischemic stroke region.

We consider a problem of nonlinear stochastic optimization with linear constraints. The method of ɛ‐feasible solution by series of Monte‐Carlo estimators has been developed for solving this problem avoiding “jamming” or “zigzagging”. Our approach is distinguished by two peculiarities: the optimality of solution is tested in a statistical manner and the Monte‐Carlo sample size is adjusted so as to decrease the total amount of Monte‐Carlo trials and, at the same time, to guarantee the estimation of the objective function with an admissible accuracy. Under some general conditions we prove by the martingale approach that the proposed method converges a.s. to the stationary point of the problem solved. As a counterexample the maximization of the probability of portfolio desired return is given, too.

A new digital signature scheme in non‐commutative Gaussian monoid is presented. Two algebraic structures are employed: Gaussian monoid and a certain module being compatible with a monoid. For both monoid and module, presentation and action level attributes are defined. Monoid action level is defined as monoid element (word) action on module element as an operator. A module is a set of functions (elements) with special properties and could be treated as some generalization of vector space.

In this article we propose a novel Wavelet Packet Decomposition (WPD)‐based modification of the classical Principal Component Analysis (PCA)‐based face recognition method. The proposed modification allows to use PCA‐based face recognition with a large number of training images and perform training much faster than using the traditional PCA‐based method. The proposed method was tested with a database containing photographies of 423 persons and achieved 82–89% first one recognition rate. These results are close to that achieved by the classical PCA‐based method (83–90%).

This paper describes a preliminary experiment in designing a Hidden Markov Model (HMM)‐based part‐of‐speech tagger for the Lithuanian language. Part‐of‐speech tagging is the problem of assigning to each word of a text the proper tag in its context of appearance. It is accomplished in two basic steps: morphological analysis and disambiguation. In this paper, we focus on the problem of disambiguation, i.e., on the problem of choosing the correct tag for each word in the context of a set of possible tags. We constructed a stochastic disambiguation algorithm, based on supervised learning techniques, to learn hidden Markov model's parameters from hand‐annotated corpora. The Viterbi algorithm is used to assign the most probable tag to each word in the text.

This paper proposes the use of a particle filter combined with color, depth information and shape features as an efficient and effective way to deal with tracking a head on the basis of image stream coming from a mobile stereovision camera. The head is modeled in the 2D image domain by an ellipse. The color distribution within interior of the ellipse is represented by a color histogram. The color histogram is dynamically updated over time. The length of the ellipse's minor axis is determined on the basis of depth information. The particles representing the candidate ellipses are weighted in each time step in respect of intensity gradient near the edge of the ellipse and matching score of the color histograms representing the interior of an ellipse surrounding the tracked object and currently analyzed one. The proposed algorithm can track a head reliably in cases of temporal occlusions as well as varying illumination conditions by dealing with multiple hypotheses for the pose. Experimental results obtained on long image sequences show the feasibility of our approach to perform tracking a head undergoing complex changes of shape and appearance against a varying background. The tracker has been evaluated in experiments consisting in face tracking with a real mobile agent.

A quick matrix multiplication algorithm is presented and evaluated on a cluster of networked workstations consisting of Pentium hosts connected together by Ethernet segments. The obtained results confirm the feasibility of using networked workstations to provide fast and low cost solutions to many computationally intensive applications such as large linear algebraic systems. The paper also presents and verifies an accurate timing model to predict the performance of the proposed algorithm on arbitrary clusters of workstations. Through this model the viability of the proposed algorithm can be revealed without the extra effort that would be needed to carry out real testing.

In this paper on basis of the results (Dyomin et al., 2003a) the structure of Shannon information amount in the joint filtering and extrapolation problem of the stochastic processes by continuous‐discrete time memory observations is investigated. For particular class of processes with applying of the general results the problem of optimal transmission over the lag channels is considered and efficiency of filtering and extrapolation receptions under transmission over channels with memory or lag is investigated.