http://sp.cs.tut.fi/publications/theses-d.shtml Theses of Institute of Signal Processing en-us http://sp.cs.tut.fi/cgi-bin/cgiwrap/spwww/thesis.cgi?id=2 Tohka, Jussi - Doctoral thesis, 2003, This thesis deals with surface extraction from noisy volumetric images, which is a common problem in medical image analysis. Due to noise, the use of a-priori information about surface topology and shape is necessary for automatic surface extraction methods. Deformable surface models can incorporate such geometric knowledge into extraction process which is restated as an energy minimization problem. A drawback of deformable models is that the formulated minimization problem is difficult to solve because of numerous local minima and a large number of variables. This difficulty may lead to sensitivity to the initialization, complicating the unsupervised use of deformable models.The main contributions of this thesis are algorithms for solving the minimization problem globally. We propose two classes of algorithms for the task, Dual surface minimization(DSM) and a hybrid of real-coded genetic algorithms and a greedy algorithm (GAGR). By global optimization of the energy of the deformable models, we are capable f reducing the initialization sensitivity of deformable surface models, and hence enabling automation of surface extraction. Moreover, these methods for global optimization do not lead to unforseeable sensitivity to values of the model parameters, another problem common with deformable models. As our second contribution, we extend a shape modeling approach for two-dimensional contours to surfaces and analytically derive a shape model for the sphere (surface). We also consider surface extraction from positron emission tomography (PET) images as an application of the deformable model based on the DSM algorithm. This task is problematic because of high noise levels in PET as compared to the contrast of the images. Our automatic method based on the proposed deformable model reliably yielded extraction results of good accuracy as compared to the imaging resolution. The success in this application demonstrates the good properties of global optimization - based deformable models for automatic surface extraction. http://sp.cs.tut.fi/cgi-bin/cgiwrap/spwww/thesis.cgi?id=3 Hong, Jae Bum - M.Sc. thesis, 17.01.2001, (no abstract) http://sp.cs.tut.fi/cgi-bin/cgiwrap/spwww/thesis.cgi?id=4 Akdemir, Özer Koray - M.Sc. thesis, 14.03.2001, (no abstract) http://sp.cs.tut.fi/cgi-bin/cgiwrap/spwww/thesis.cgi?id=5 Hariharan, Ramalingam - Doctoral thesis, 14.03.2001, (no abstract) http://sp.cs.tut.fi/cgi-bin/cgiwrap/spwww/thesis.cgi?id=6 Partanen, Pasi - M.Sc. thesis, 14.03.2001, (no abstract) http://sp.cs.tut.fi/cgi-bin/cgiwrap/spwww/thesis.cgi?id=7 Virtanen, Tuomas - M.Sc. thesis, 14.03.2001, In audio signal spectrum modeling, the aim is to transform a signal to a more easily applicable form, removing the information that is irrelevant in signal perception. Sinusoids plus noise model is a spectral model, in which the periodic components of the sound are represented with sinusoids with time-varying frequencies, amplitudes and phases. The remaining non-periodic components are represented with a filtered noise. The sinusoidal model utilizes the physical properties of musical instruments and the noise model the humans' inability to perceive the exact spectral shape or phase of stochastic signals. In the case of polyphonic music signals, the estimation of the parameters of sinusoids is a difficult task, since the periodic components are usually not stable. A sufficient time and frequency resolution is also difficult to achieve at the same time. A big part of this thesis discusses the detection and parameter estimation of periodic components with several algorithms. In addition to already existing algorithms, a new iterative algorithm is presented, which is based on the fusion of closely spaced sinusoids. The sinusoidal model is applied in the separation of overlapping sounds and manipulation. In the sound separation, a new perceptual distance measure between sinusoids is used. The perceptual distance measure is based on the humans' way to associate spectral components into sound sources. Also a new separation method based on the multipitch estimation is explained. The modification of the pitch and time scale of sounds with the sinusoid plus noise model without affecting the quality of the sound is explained shortly, too. http://sp.cs.tut.fi/cgi-bin/cgiwrap/spwww/thesis.cgi?id=8 Lundan, Miikka - M.Sc. thesis, 11.04.2001, (no abstract) http://sp.cs.tut.fi/cgi-bin/cgiwrap/spwww/thesis.cgi?id=9 Kiss, Imre - Doctoral thesis, 09.05.2001, (no abstract) http://sp.cs.tut.fi/cgi-bin/cgiwrap/spwww/thesis.cgi?id=10 Nieminen, Toni - M.Sc. thesis, 09.05.2001, (no abstract) http://sp.cs.tut.fi/cgi-bin/cgiwrap/spwww/thesis.cgi?id=11 Hömmö, Esa Henrik - M.Sc. thesis, 06.06.2001, (no abstract) http://sp.cs.tut.fi/cgi-bin/cgiwrap/spwww/thesis.cgi?id=12 Mäkinen, Jari - M.Sc. thesis, 06.06.2001, (no abstract) http://sp.cs.tut.fi/cgi-bin/cgiwrap/spwww/thesis.cgi?id=13 Giurcaneanu, Ciprian Doru - Doctoral thesis, 22.08.2001, (no abstract) http://sp.cs.tut.fi/cgi-bin/cgiwrap/spwww/thesis.cgi?id=14 Iordache, Razvan - Doctoral thesis, 22.08.2001, (no abstract) http://sp.cs.tut.fi/cgi-bin/cgiwrap/spwww/thesis.cgi?id=15 Lähdesmäki, Harri - M.Sc. thesis, 22.08.2001, This thesis is carried out in the Institute of Signal Processing at Tampere University of Technology. The research is a part of a larger interdisciplinary project where archeologically important cave inscriptions are to be preserved and studied. The project consists of three parts: extraction and storage of the inscriptions, post-processing of the extracted data, and study of the historical petroglyphs. The focus of this thesis is in the post-processing of the extracted topographical data using image processing tools. The goal of this work is not only to produce topographic maps as accurately as possible, e.g. by removing the special types of noise, but rather to make the inscriptions more visible and easier to read. Proper noise removal methods, however, are designed to get rid of the noise, and for making the later processing easier and more reliable. Characteristics of the noise are also estimated during the design of the noise removal methods. Two conceptually different approaches have been considered for improving the legibility of the inscriptions: traditional enhancement and segmentation. Loosely speaking, an enhancement method is a numerical algorithm that manipulates the image in a way such that the selected features are either emphasized or suppressed, without introducing any detoriarating or artificial effects. Unsharp masking, histogram based, and special nonlinear filtering methods are examined and adjusted for enhancement purposes. Segmentation methods, on the other hand, try to segment the image into its constituent objects, in our case, into the inscriptions and the background. Different feature extraction methods, like edge detectors and local activity indicators, are employed for aiding the final segmentation process. Finally, special thresholding and morphological operators are designed to conclude the segmentation. The simulation results of different algorithms on sample cave inscription images are presented. Various methods performed differently, each of them having some desirable properties and drawbacks. Therefore, the selection of the best methods is left for archeologists who aim to study these inscriptions. http://sp.cs.tut.fi/cgi-bin/cgiwrap/spwww/thesis.cgi?id=16 Peltonen, Vesa - M.Sc. thesis, 22.08.2001, (no abstract) http://sp.cs.tut.fi/cgi-bin/cgiwrap/spwww/thesis.cgi?id=17 Rämö, Pauli - M.Sc. thesis, 22.08.2001, (no abstract) http://sp.cs.tut.fi/cgi-bin/cgiwrap/spwww/thesis.cgi?id=18 Alban Bucheli, Edisson - M.Sc. thesis, 10.10.2001, (no abstract) http://sp.cs.tut.fi/cgi-bin/cgiwrap/spwww/thesis.cgi?id=19 Arian, Peyman - M.Sc. thesis, 10.10.2001, (no abstract) http://sp.cs.tut.fi/cgi-bin/cgiwrap/spwww/thesis.cgi?id=20 Eronen, Antti Johannes - M.Sc. thesis, 10.10.2001, (no abstract) http://sp.cs.tut.fi/cgi-bin/cgiwrap/spwww/thesis.cgi?id=21 Kesseli, Juha - M.Sc. thesis, 10.10.2001, (no abstract) http://sp.cs.tut.fi/cgi-bin/cgiwrap/spwww/thesis.cgi?id=22 Reijonen, Pasi - M.Sc. thesis, 10.10.2001, (no abstract) http://sp.cs.tut.fi/cgi-bin/cgiwrap/spwww/thesis.cgi?id=23 Linne, Marja-Leena - Doctoral thesis, 14.11.2001, (no abstract) http://sp.cs.tut.fi/cgi-bin/cgiwrap/spwww/thesis.cgi?id=24 Martikainen, Kalle - M.Sc. thesis, 14.11.2001, (no abstract) http://sp.cs.tut.fi/cgi-bin/cgiwrap/spwww/thesis.cgi?id=25 Mattila, Ville-Veikko - Doctoral thesis, 14.11.2001, (no abstract) http://sp.cs.tut.fi/cgi-bin/cgiwrap/spwww/thesis.cgi?id=26 Seppänen, Jarno - M.Sc. thesis, 14.11.2001, (no abstract) http://sp.cs.tut.fi/cgi-bin/cgiwrap/spwww/thesis.cgi?id=27 Gonzalez, Santiago - M.Sc. thesis, 12.12.2001, (no abstract) http://sp.cs.tut.fi/cgi-bin/cgiwrap/spwww/thesis.cgi?id=28 Nikkanen, Jarno - M.Sc. thesis, 12.12.2001, (no abstract) http://sp.cs.tut.fi/cgi-bin/cgiwrap/spwww/thesis.cgi?id=29 Salmi, Jussi - M.Sc. thesis, 12.12.2001, (no abstract) http://sp.cs.tut.fi/cgi-bin/cgiwrap/spwww/thesis.cgi?id=30 Tico, Marius - Doctoral thesis, 12.12.2001, (no abstract) http://sp.cs.tut.fi/cgi-bin/cgiwrap/spwww/thesis.cgi?id=31 Wang, Ye - Doctoral thesis, 16.01.2002, (no abstract) http://sp.cs.tut.fi/cgi-bin/cgiwrap/spwww/thesis.cgi?id=32 Medina Trujillo, Juan - M.Sc. thesis, 13.02.2002, (no abstract) http://sp.cs.tut.fi/cgi-bin/cgiwrap/spwww/thesis.cgi?id=33 Nykter, Matti - M.Sc. thesis, 13.03.2002, (no abstract) http://sp.cs.tut.fi/cgi-bin/cgiwrap/spwww/thesis.cgi?id=34 Tuori, Miikka - M.Sc. thesis, 13.03.2002, (no abstract) http://sp.cs.tut.fi/cgi-bin/cgiwrap/spwww/thesis.cgi?id=35 Niemistö, Antti - M.Sc. thesis, 10.04.2002, Modern remote sensing systems provide images that correspond to Earth terrains with areas of thousands of square kilometers. These images offer the user the opportunity to carry out monitoring of various physical phenomena and of various scenes. Usually this requires that some classification technique is applied to each obtained image. Due to the large amount of pixels in the images, it is not realistic to expect a human expert to classify them reliably by means of visual analysis. Automatic classification of remote sensing images is therefore a key problem. A number of different techniques for local recognition of remote sensing images have been proposed. However, none of the techniques is able to classify the images without misclassifications. The main reason for this is the influence of noise that is inherently present in remote sensing images. In this thesis a technique for correcting these misclassifications is developed. The technique is based on local nonlinear operations defined by a weighted graph. The optimal weight function for the graph is obtained using genetic algorithms in a training-based optimization framework. Finally, the applicability of the optimal operator to images on which it is not trained is verified. http://sp.cs.tut.fi/cgi-bin/cgiwrap/spwww/thesis.cgi?id=36 Toivonen, Iiro - M.Sc. thesis, 10.04.2002, (no abstract) http://sp.cs.tut.fi/cgi-bin/cgiwrap/spwww/thesis.cgi?id=37 Friis, Soren - M.Sc. thesis, 15.05.2002, (no abstract) http://sp.cs.tut.fi/cgi-bin/cgiwrap/spwww/thesis.cgi?id=38 Hourunranta, Ari - Licensiate thesis, 05.06.2002, (no abstract) http://sp.cs.tut.fi/cgi-bin/cgiwrap/spwww/thesis.cgi?id=39 Korhonen, Saku - M.Sc. thesis, 05.06.2002, (no abstract) http://sp.cs.tut.fi/cgi-bin/cgiwrap/spwww/thesis.cgi?id=40 Pasanen, Antti - M.Sc. thesis, 05.06.2002, (no abstract) http://sp.cs.tut.fi/cgi-bin/cgiwrap/spwww/thesis.cgi?id=41 Pirinen, Tuomo - M.Sc. thesis, 05.06.2002, (no abstract) http://sp.cs.tut.fi/cgi-bin/cgiwrap/spwww/thesis.cgi?id=42 Sastry Madugandur, Prakash - M.Sc. thesis, 05.06.2002, (no abstract) http://sp.cs.tut.fi/cgi-bin/cgiwrap/spwww/thesis.cgi?id=43 Varjos, Mikko - M.Sc. thesis, 05.06.2002, (no abstract) http://sp.cs.tut.fi/cgi-bin/cgiwrap/spwww/thesis.cgi?id=44 Burian, Adrian - Doctoral thesis, 21.08.2002, (no abstract) http://sp.cs.tut.fi/cgi-bin/cgiwrap/spwww/thesis.cgi?id=45 Heikkinen, Ari - Doctoral thesis, 21.08.2002, (no abstract) http://sp.cs.tut.fi/cgi-bin/cgiwrap/spwww/thesis.cgi?id=46 Yli-Kaakinen, Juha - Doctoral thesis, 21.08.2002, This thesis concentrates on optimizing both infinite-impulse response(IIR) and finite-impulse response (FIR) digital filters for variouspractical applications. The main emphasis is laid on designing filterssuch that all the coefficients are representable using few powers oftwo. For these purposes, various efficient strategies are developedfor optimizing both infinite-precision and finite-precision filterssubject to the given constraints. For designing infinite-precision IIRfilters a two-step approach is proposed. First, a sub-optimum solutionis generated using a simple systematic algorithm. Second, thisstart-up solution is further improved using an efficientgeneral-purpose nonlinear optimization algorithm, giving the optimalsolution. The desired finite-precision coefficients can be convenientlydetermined using the following two steps. First, an optimizationalgorithm is used for determining a parameter space of theinfinite-precision coefficients including the feasible space where thefilter meets the given criteria. The second step then involvesfinding the filter parameters in this space such that the resultingfilter meets the given criteria with the simplest coefficientrepresentation forms. The proposed algorithm guarantees that theoptimum multiplierless finite-wordlength solution can be found. Inthe case of IIR filters, the first step is carried out by properlyutilizing the above-mentioned two-step approach, whereas in the caseof linear-phase FIR filters, linear programming is applied. In addition to generating efficient optimization techniques forcoefficient quantization, a very efficient structure is introduced forimplementing IIR filters based on the cascade connection of low-orderlattice wave digital (LWD) filters. Compared to the conventionalimplementation, this structure requires significantly lower data andcoefficient wordlengths. The proposed optimization techniques are also applied to improving theperformance of pipelined IIR digital filters and to optimizingfinite-precision conventional and approximately linear-phase LWDfilters as well as finite-precision FIR digital filters. Several examples are included illustrating the efficiency andflexibility of the proposed algorithms compared to other existingtechniques. http://sp.cs.tut.fi/cgi-bin/cgiwrap/spwww/thesis.cgi?id=47 Zhang, Yihong - M.Sc. thesis, 21.08.2002, (no abstract) http://sp.cs.tut.fi/cgi-bin/cgiwrap/spwww/thesis.cgi?id=48 Luoma, Jouni - M.Sc. thesis, 18.09.2002, (no abstract) http://sp.cs.tut.fi/cgi-bin/cgiwrap/spwww/thesis.cgi?id=49 Happonen, Antti - M.Sc. thesis, 16.10.2002, (no abstract) http://sp.cs.tut.fi/cgi-bin/cgiwrap/spwww/thesis.cgi?id=50 Lamminmäki, Elina - M.Sc. thesis, 13.11.2002, (no abstract) http://sp.cs.tut.fi/cgi-bin/cgiwrap/spwww/thesis.cgi?id=51 Sillanpää, Maria - M.Sc. thesis, 13.11.2002, (no abstract) http://sp.cs.tut.fi/cgi-bin/cgiwrap/spwww/thesis.cgi?id=52 Gutierrez, Amelia - M.Sc. thesis, 11.12.2002, (no abstract) http://sp.cs.tut.fi/cgi-bin/cgiwrap/spwww/thesis.cgi?id=53 Kuja-Halkola, Sami - M.Sc. thesis, 11.12.2002, (no abstract) http://sp.cs.tut.fi/cgi-bin/cgiwrap/spwww/thesis.cgi?id=54 Mattila, Heikki - M.Sc. thesis, 11.12.2002, (no abstract) http://sp.cs.tut.fi/cgi-bin/cgiwrap/spwww/thesis.cgi?id=55 Partio, Mari - M.Sc. thesis, 11.12.2002, Due to rapid increase in volume of image and video collections, traditional methods of indexing and retrieval using only keywords have become outdated. Therefore, alternative methods to describe images using their visual content have been developed. To produce and test algorithms for content-based image and video retrieval, MUVIS (Multimedia Video Indexing and Retrieval System) was developed at TUT. The goal of MUVIS is a fast, real-time and reliable audio/video (AV) browsing and indexing application, which is also capable of extracting some key features (such as color, texture and shape) of the AV media. Most of the existing image retrieval systems perform reasonably when using color features. However, retrieval accuracy using shape or texture features does not produce as good results. Therefore, this thesis investigates different methods of representing shape and texture in content-based image retrieval. Later, when appropriate segmentation algorithms are available some of these methods could also be applied to video object retrieval. The thesis presents two contributions: one is shape-based and the second is texture-based retrieval method. The former contribution concerns shape analysis and retrieval. Shape attributes can be roughly divided into two main categories: boundary-based and region-based. Since the human visual system itself focuses on edges and ignores uniform regions, this thesis concentrates on boundary-based representations. A novel boundary-based method using distance transformation and ordinal correlation is developed in this thesis. Simulation results show that the proposed technique produced encouraging results when using MPEG-7 shape test database. The second contribution of the thesis is a constrained application in which the database contains a set of rock images. In this application, we applied a technique based on Gray-Level Co-occurrence matrices (GLCM) and compared the results with a well-known method from the literature. It was found that GLCM outperforms Gabor Wavelet features when considering retrieval time and visual quality of the results. http://sp.cs.tut.fi/cgi-bin/cgiwrap/spwww/thesis.cgi?id=56 Rakkolainen, Ismo - Doctoral thesis, 11.12.2002, The main objective of this thesis is to present novel applications, ideas, methods, and insights for virtual reality and some related fields of technology. Virtual reality and 3D graphics will impact the digital future in numerous ways. The thesis is based on several papers and patent applications. The thesis consists of three major present and future domains of virtual reality. One area is the work done by the author and his research group on building new applications with networked 3D graphics (Web3D). The implemented work on visualizing real-time position data and on aiding navigation and service finding in real places with Web3D is presented. The work has been application-oriented, using standards as much as possible. Several of the results are the first of their kind. The results may contribute to the development of mobile 3D applications. Some novel 3D modeling concepts are also presented. The second area is novel user interfaces. The author presents a novel approach for mobile user interfaces, which is largely derived from virtual reality. The user interface concept can be applied to numerous future devices. The work is still in its early stage. The third area of research is some novel methods for immersive projection technology. The main contribution is a new kind of a penetrable, non-solid fog screen. It enables high-quality walk-thru virtual environments, new projection experiences, and also practically removes the risk of screen fragility. Some technical extensions for the fog screen are envisioned. Also a novel, so far unimplemented tracking screen concept is presented. http://sp.cs.tut.fi/cgi-bin/cgiwrap/spwww/thesis.cgi?id=57 Rehtonen, Petri - M.Sc. thesis, 11.12.2002, (no abstract) http://sp.cs.tut.fi/cgi-bin/cgiwrap/spwww/thesis.cgi?id=58 Rustari, Mika - M.Sc. thesis, 11.12.2002, (no abstract) http://sp.cs.tut.fi/cgi-bin/cgiwrap/spwww/thesis.cgi?id=59 Similä, Maisa - M.Sc. thesis, 11.12.2002, (no abstract) http://sp.cs.tut.fi/cgi-bin/cgiwrap/spwww/thesis.cgi?id=60 Tsoumanis, Andreas - M.Sc. thesis, 11.12.2002, (no abstract) http://sp.cs.tut.fi/cgi-bin/cgiwrap/spwww/thesis.cgi?id=61 Khriji, Lazar - Doctoral thesis, 20.12.2002, This thesis deals with various design and implementation aspects of a new class of nonlinear rational-based filters. A rational function, defined as the ratio of two polynomials, offers the following advantages. First, it is a universal approximator and a good extrapolator. Second, it can be trained by a linear adaptive algorithm, which makes it suitable for real-time adaptive signal/image processing. Third, it can best approximate some special functions. Rational filters have been shown to be a good alternative for many digital signal processing applications where linear filters can not provide satisfactory solutions, e.g. digital image processing. In order to demonstrate its utility as a tool for solving signal and image processing problems, rational-based filters are applied to three main problems: interpolation; enhancement; and restoration. The first contribution of the thesis is the introduction of new rational interpolation algorithms. The proposed methods for image interpolation are able to accurately reconstruct images yielding both sharp details and low blocking artifacts. Furthermore, in order to deal with multichannel signals, the concept of rational filtering is extended to vector-valued signals by defining the vector rational filter. As opposed to component-wise processing, the vector rational filter processes signal samples as vectors, inherently utilizing the dependencies between signal components. As a consequence, the corresponding algorithms exhibit desirable properties, such as, edge and detail preservation and accurate chromaticity estimation. An inherent problem in most image enhancement schemes is the amplification of noise, which, due to Weber's law, is mostly visible in the darker portions of an image. Using a newly modified Unsharp Masking technique, one can adapt the enhancement process in a computationally efficient way. This is achieved by introducing a control term expressed as a rational function of the local input data. In this way and under certain assumptions, noise amplification is avoided and, at the same time, overshoot effects on sharp edges are limited. A new class of nonlinear hybrid filters, the median-rational hybrid filters (MRHF), is introduced with different structures for signal and image restoration. The noise attenuation and detail preservation capabilities of these filters are investigated. To deal with multichannel signals, the MRHF filters are extended to vector-valued signals. The effectiveness of the vector median-rational hybrid filters (VMRHF) for removing additive noise with different scenarios from color images is demonstrated by means of computer simulations. For the purpose of image restoration, an adaptive class of nonlinear multi- channel image processing filters is developed. The filters use in an effective way, fuzzy memberships, average filters and distance criteria. They combine the good behavior of rational functions and fuzzy logic with vector approach. The design of such a filter is based on fuzzy logic to adapt to local data in such a way as to allow characteristics of the human visual system to be considered. The performance of the developed filters is accessed through computer simulations with different noise distributions. In addition, a dedicated hardware system for the MRHF is developed, exploiting the features of the dynamic logic families, which can achieve high speed and compactness, while keeping the power dissipation very low. The resulting structure is suitable for real time applications where size and cost are of critical significance. http://sp.cs.tut.fi/cgi-bin/cgiwrap/spwww/thesis.cgi?id=62 Guldogan, Esin - M.Sc. thesis, 15.01.2003, (no abstract) http://sp.cs.tut.fi/cgi-bin/cgiwrap/spwww/thesis.cgi?id=63 Rantanen, Pasi - M.Sc. thesis, 15.01.2003, (no abstract) http://sp.cs.tut.fi/cgi-bin/cgiwrap/spwww/thesis.cgi?id=64 Heine, Kari - M.Sc. thesis, 12.03.2003, (no abstract) http://sp.cs.tut.fi/cgi-bin/cgiwrap/spwww/thesis.cgi?id=65 Niemistö, Riitta - Doctoral thesis, 12.03.2003, This thesis proposes algorithms to be utilized for iterative design of linear filters and for adaptive polynomial filtering. The goal is to improve existing algorithms in terms of filtering performance, time and complexity requirements. First, a family of fast algorithms suitable for adaptive polynomial filtering, namely LS-LMS algorithm and its square root variants, is introduced and its connection with the family of RLS algorithms is analyzed. The algorithms are derived by combining the triangularization and backsolving steps of QR-RLS algorithm. In the derivation we allow a modification of the cost function in order to decrease the complexity of the algorithm from O(M^2) to O(M), where M is the number of filter parameters. Second, adaptive filtering algorithms are proposed for the problem of acoustic echo control in the case of strong acoustic distortion in the loudspeaker that decreases the echo attenuation obtained using adaptive linear filtering algorithms in echo cancelers. Several existing algorithms were tested using both linear and homogeneous Volterra filters, but their performance was not found satisfactory either in terms of performance or complexity. A polynomial preprocessor followed by a linear filter was found to perform well when the preprocessor was adapted by the robust, but computationally demanding, QR-RLS algorithm and the linear part was adapted by the normalized LMS algorithm. Third, the problem of optimum energy compaction for FIR and IIR filters is addressed. The analytical method for optimum compaction FIR filter design is analyzed and shown to provide a very fast algorithm under certain conditions. For the design of optimum compaction IIR filters two new methods are proposed. The first design method operates with the numerator and the denominator in the causal part of the associated product filter, which are in turn optimized via iterative relaxations. The second method is similar, except that the angles of the poles are set to fixed values, situated at the transitions in the frequency representation of the corresponding ideal brickwall filter. Thus, the second method is faster than the first, and, moreover, it was observed that when the ideal brickwall filter does not have too many transitions, the second method provides filters with better performance. Both methods make use of semidefinite programming optimization using an appropriate parameterization of positive real polynomials. The last part of the thesis provides contributions to the frequency selective IIR filter design. The frequency response of the filter is fitted to an ideal frequency response by optimizing either a weighted sum of magnitude squares (least squares) criterion or a minimax (Chebyshev) criterion. For the least squares criterion a new convex stability domain is used and experimental evidence is given that the best designs are usually obtained with a multistage algorithm where three methods, Steiglitz-McBride, Gauss-Newton and classical descent, are used consecutively in that order. The result of multistage least squares IIR design is used to initialize the optimization procedure based on a Chebyshev criterion, for which a simplified iterative procedure is proposed. In the simplified procedure only the numerator is updated after initialization. The simplified procedure is compared experimentally with the complete procedure and the results show only very small differences in the criterion. http://sp.cs.tut.fi/cgi-bin/cgiwrap/spwww/thesis.cgi?id=66 Ala-Kleemola, Timo - M.Sc. thesis, 09.04.2003, (no abstract) http://sp.cs.tut.fi/cgi-bin/cgiwrap/spwww/thesis.cgi?id=67 Mäkinen, Jarno - M.Sc. thesis, 09.04.2003, (no abstract) http://sp.cs.tut.fi/cgi-bin/cgiwrap/spwww/thesis.cgi?id=68 Rantala, Mikko - M.Sc. thesis, 09.04.2003, (no abstract) http://sp.cs.tut.fi/cgi-bin/cgiwrap/spwww/thesis.cgi?id=69 Vesanen, Tomi - M.Sc. thesis, 09.04.2003, (no abstract) http://sp.cs.tut.fi/cgi-bin/cgiwrap/spwww/thesis.cgi?id=70 Lugmayr, Artur Roland - Licensiate thesis, 07.05.2003, (no abstract) http://sp.cs.tut.fi/cgi-bin/cgiwrap/spwww/thesis.cgi?id=71 Parviainen, Mikko - M.Sc. thesis, 07.05.2003, (no abstract) http://sp.cs.tut.fi/cgi-bin/cgiwrap/spwww/thesis.cgi?id=72 San Jose Garcia, Jorge - M.Sc. thesis, 07.05.2003, (no abstract) http://sp.cs.tut.fi/cgi-bin/cgiwrap/spwww/thesis.cgi?id=73 Viitaniemi, Timo - M.Sc. thesis, 07.05.2003, (no abstract) http://sp.cs.tut.fi/cgi-bin/cgiwrap/spwww/thesis.cgi?id=74 Ahmed, Iftikhar - Licensiate thesis, 04.06.2003, (no abstract) http://sp.cs.tut.fi/cgi-bin/cgiwrap/spwww/thesis.cgi?id=75 Aho, Tommi - M.Sc. thesis, 04.06.2003, This thesis is a description of the simulation tool for genetic regulatory networks (GRN). The thesis is a part of an interdisciplinary project, which is entitled "Dynette," and which has been carried out jointly by the Institute of Signal Processing in Tampere University of Technology and MediCel Ltd., Helsinki. The overall purpose of the project has been the development simulation and modeling tools for genetic regulatory networks. The project has been financially supported by Tekes, the National Technology Agency of Finland. The GRNs are constructed by numerous interactions between different types of molecules. The genetic code of the genome is able to regulate many of the fundamental processes of cells by utilizing the GRNs, for instance energy production and metabolism. Therefore, the GRNs are considered very essential for the understanding of the vital functions and the diseases of organisms. The novel measurement techniques of molecular biology can be used for producing large amounts of data that describe the functions of GRNs. Systems biology utilizes the data in mathematical modeling and simulation. This thesis first reviews the mathematical methods that are presented in the literature. Based on these studies, the Simulation Tool has been implemented in which concentration time-series can be described using separate mathematical models for each molecule. A network model describing the behavior of the studied GRN can be constructed by combining these individual models together. The computational experiments are performed using the data which describes the behavior of a specific GRN in yeast cells. An indefinite amount of noiseless data for that GRN can be produced using a mathematical model proposed in the literature. The behavior of the studied GRN can be simulated with the other models generated by the Simulation Tool. However, none of the implemented model types can simulate the behavior of all molecule types. In this work, it is demonstrated that if the models of the molecules are identified separately from each other during the learning process, the simulation of the network model may not behave in a satisfactory manner. We conclude that the Simulation Tool can be used for studying biochemical time-series data. The results of this thesis direct the further development of the Simulation Tool and the associated mathematical modeling methods. http://sp.cs.tut.fi/cgi-bin/cgiwrap/spwww/thesis.cgi?id=76 Barna, Laurentiu - M.Sc. thesis, 04.06.2003, (no abstract) http://sp.cs.tut.fi/cgi-bin/cgiwrap/spwww/thesis.cgi?id=77 Gomez Herrero, German - M.Sc. thesis, 04.06.2003, The mismatch negativity (MMN) is an event-related potential elicited by changes in repetitive auditory stimuli, irrespective of the direction of the subject's attention. It is mainly generated in the auditory cortex suggesting the ability of the brain to automatically perform complex comparisons between new sounds and the immediate auditory past. This pre-perceptual processing in the auditory cortex tends to trigger frontal cortex activity probably representing the initiation of attention switch to sound change. The MMN is, nowadays, the only objective measure of the accuracy of central auditory processing. Furthermore, it has shown its utility in measuring higher level brain processing tasks related to the attention switch to changes in the acoustic environment. One of the most promising fields of application of the MMN involves newborns and young infants since it has been shown that the MMN is elicited even in prematurely born newborns. A fundamental problem found in MMN research is the very low signal to noise ratios that are obtained using classical EEG recording techniques. The MMN component, whose amplitude is typically in the range of 1-3 microvolts, is embedded in the ongoing EEG activity that can reach peak variations of several hundreds of microvolts. During the recent years, several studies have shown the utility of wavelets in biomedical signal processing. In this Thesis we present two approaches based on wavelets to detect the presence of MMN deflection in single ERP trials. The first approach is based on classification, using a neural network, of the wavelet features representing the time-frequency window of the MMN component. Results obtained with such approach have shown the utility of the procedure in selecting the single ERP trials with stronger MMN deflection in order to generate selective averages with improved signal to noise ratios. However, the low classification ratios suggested that wavelet features are not enough to effectively characterize the MMN due to its overlap with many other EEG components in the time-frequency plane. The second approach tries to overcome this limitation by combining wavelet features with Independent Component Analysis (ICA). In many cases, ICA was able to separate the MMN from the background activity making the wavelet representation of the MMN much more accurate and therefore, yielding better classification ratios of single trials with MMN deflection. http://sp.cs.tut.fi/cgi-bin/cgiwrap/spwww/thesis.cgi?id=78 Kabara, Sid Ahmed - M.Sc. thesis, 04.06.2003, (no abstract) http://sp.cs.tut.fi/cgi-bin/cgiwrap/spwww/thesis.cgi?id=79 Lehtinen, Vesa - M.Sc. thesis, 04.06.2003, (no abstract) http://sp.cs.tut.fi/cgi-bin/cgiwrap/spwww/thesis.cgi?id=80 Niiranen, Samuli - Licensiate thesis, 04.06.2003, (no abstract) http://sp.cs.tut.fi/cgi-bin/cgiwrap/spwww/thesis.cgi?id=81 Öktem, Hakan - Doctoral thesis, 04.06.2003, (no abstract) http://sp.cs.tut.fi/cgi-bin/cgiwrap/spwww/thesis.cgi?id=82 Pertilä, Pasi - M.Sc. thesis, 04.06.2003, (no abstract) http://sp.cs.tut.fi/cgi-bin/cgiwrap/spwww/thesis.cgi?id=83 Shahed hagh ghadam, Ali - M.Sc. thesis, 04.06.2003, (no abstract) http://sp.cs.tut.fi/cgi-bin/cgiwrap/spwww/thesis.cgi?id=84 Soria-Rodriguez, Marcelo - M.Sc. thesis, 04.06.2003, (no abstract) http://sp.cs.tut.fi/cgi-bin/cgiwrap/spwww/thesis.cgi?id=85 Bregovic, Robert - Doctoral thesis, 20.08.2003, By using multirate filter banks, the signal under consideration can be separated in the frequency domain into two or more subband signals, each of them containing only one spectral part of the original signal. When applying various signal processing algorithms to these subbands, such as compression or denoising, significantly better performances may be achieved compared to the case where the same algorithms are applied to the original signal itself. Therefore, many modern digital signal processing algorithms are based on processing subband signals instead of directly processing the original signal. The crucial elements for this process are multirate filter banks. This thesis concentrates mainly on synthesizing multirate filter banks of two types, namely, alias-free two-channel filter banks and multi-channel (M-channel) filter banks. In the case of two-channel filter banks, the filter banks are first classified into various types and then design methods for filter banks based on the use of finite-impulse response (FIR) and infinite-impulse response (IIR) filters are proposed. The main emphasis is placed on methods for designing filter banks by minimizing the maximum of the stopband energies (least squares design) of the two analysis filters subject to the given passband and transition-band constraints and the given allowable reconstruction error. The proposed methods can be easily extended for minimizing the stopband ripples (minimax design) of the filters in the bank. Additionally, the properties of FIR filter banks designed by using the peak-constrained least squares optimization criterion are studied. This criterion provides a tradeoff between the least squares and the minimax criteria, which is very useful in some applications. In the case of M-channel filter banks, design methods for orthogonal (linear-phase) and biorthogonal (low-delay) cosine-modulated filter banks are proposed. The main emphasis is placed on cosine-modulated filter banks due to their efficient implementations and drastically reduced design complexity (only one prototype filter has to be designed) when compared to the direct design of M-channel filter banks. The design of perfect-reconstruction (PR) orthogonal filter banks can be further simplified by optimizing the angles of the lattice structures for the prototype filter instead of the prototype filter coefficients itself. Most of the methods introduced for designing prototype filters for cosine-modulated filter banks are based on a multi-step design method. The design starts with a two-channel case, and then, in each step the number of channels is increased. The result of the previous step is modified and used as a start-up solution for the next step. A noted exception is the design of PR biorthogonal filter banks that uses a state-of-the-art optimization technique called the second-order cone programming. For solving the constrained optimization problems that arise in the design procedures for two-channel and M-channel filter banks, a modified version of the Dutta-Vidyasagar algorithm is used. In addition to the optimization-based methods, some iterative methods are introduced. When designing filter banks, the same emphasis is placed on PR and nearly perfect-reconstruction (NPR) filter banks. It has been demonstrated that for most practical applications, the PR property is not needed. By relaxing the PR property, filter banks having better properties can be obtained, such as higher channel selectivity or a shorter filter bank delay. Alternatively, the same overall performance compared to PR filter banks is achievable by using shorter filters in the bank. Many examples are included illustrating the efficiency and flexibility of the proposed design methods. http://sp.cs.tut.fi/cgi-bin/cgiwrap/spwww/thesis.cgi?id=86 Lukkarinen, Teemu - M.Sc. thesis, 20.08.2003, (no abstract) http://sp.cs.tut.fi/cgi-bin/cgiwrap/spwww/thesis.cgi?id=87 Schorpp, Marcus - Licensiate thesis, 20.08.2003, (no abstract) http://sp.cs.tut.fi/cgi-bin/cgiwrap/spwww/thesis.cgi?id=88 Vasilache, Adriana - Doctoral thesis, 20.08.2003, (no abstract) http://sp.cs.tut.fi/cgi-bin/cgiwrap/spwww/thesis.cgi?id=89 Vihonen, Juho - M.Sc. thesis, 20.08.2003, Pulssidopplertutkia käytetään laajastisiviili-ja sotilassovelluksissa. Tällaiset tutkat pyrkivät havaitsemaan tietyllänopeudella liikkuvia kohteita. Tässätyössä esitellään aihepiiriin liittyviä käsitteitä, mutta pääpaino on kuitenkin kohteen havaitsemiseen liittyvissä tekniikoissa. Sen lisäksi työssä esitellään kohteen havaitsemiseen käytettävää uutta tekniikkaa. Uuden menetelmän hyvä soveltuvuus kohteen ilmaisuun osoitetaan simulaatioilla. Lukijalta odotetaan tiettyä kokemusta tutkatekniikan aihepiiristä.Tarkastelu keskittyy pääsääntöisesti ilma-alustalla toimivaan pulssitutkaan ja erityisesti kohteen havaitsemisen maata vasten mahdollistavaan moodiin. Lähtöoletuksena on, että tutka toimii X-alueella useiden kHz:n pulssintoistotaajuudella. On täysin perusteltua väittää, että tutka on tänä päivänä varsin monimutkainen laite. Pulssidopplertutkan toiminnan hallitseminen vaatiikin perusteelliset tiedot lukuisilta eri osa-alueilta. Pääsääntöisesti tässä työssä keskitytään digitaaliseen signaalinkäsittelyyn aika-ja taajuustasossa. Käsiteltävätasiat, kuten aaltomuodon optimointi ja kohteiden havaitseminen, vaativat lisäksi esimerkiksi optimointiteorian ja todennäköisyysmatematiikan hallintaa. Kokonaisuutena työ tähtää tutkan suorituskyvyn parantamiseen täysin toteutettavissa olevilla signaalinkäsittelymenetelmillä. http://sp.cs.tut.fi/cgi-bin/cgiwrap/spwww/thesis.cgi?id=90 Viljakainen, Marko - M.Sc. thesis, 20.08.2003, (no abstract) http://sp.cs.tut.fi/cgi-bin/cgiwrap/spwww/thesis.cgi?id=91 Gotchev, Atanas - Doctoral thesis, 17.09.2003, One has to solve two main problems when trying to transfer real-world phenomena into sequences of binary digits to be further processed and analyzed. First, there is the problem of defining the decomposition signal space through an appropriate basis. This quite abstract geometrical approach considers signal processing as a kind of projection of the signal on the space thus defined. The result is a set of elements called ‘model coefficients’, ‘coordinates’, ‘spectral lines’, ‘features’, etc. that helps to classify the signal and to say: ‘Yes, this signal is important ‘here…’; ‘This signal is ‘close’ to this pattern…’; ‘These signals can be treated in the following manner…’, etc. The second problem is to design efficient algorithms, preferably in the form of filter structures, to make the abstract constructions operational. Splines and wavelets are good solutions of the above-mentioned two problems. In particular, their success in many areas, such as resizing, compression, denoising, analysis, etc. is due to the fact that they yield signal decompositions in multiple scales and resolutions and help to reveal signal features of different sizes and different time, space, or frequency localizations. This thesis takes advantage of multiscale bases and investigates for better answers to some problems of practical importance. The achievements of the thesis can be broadly separated into two parts. In the first part, the basic scale-changing operators, i.e. interpolators and decimators, aimed at arbitrary scale image resizing are improved for their performance and speed. Use is made of the class of piecewise basis functions of minimal support constructed as combinations of uniform B-splines of different degrees. They are very susceptible for optimization while being also very efficient in realization. The adjust-able parameters in the combinations are tuned by a minimax optimization technique specified in the Fourier domain. The optimized interpolators show very good performance both in experiments with real images and in terms of error kernel behavior. Specifically, the error kernel keeps very good shape for high frequencies close to the Nyquist frequency, while for low frequencies the error is very small and controllable through the optimization procedure. Hence, the interpolators thus optimized are especially appropriate for high-frequency images, i.e. images containing sharp edges and small details. Based on the designed optimized basis functions, a new near least squares decimation operator is proposed as a compromise between the L2 and l2 norm minimization cases. This proposal helps to derive a computationally efficient decimation structure possessing good anti-aliasing properties. It considerably reduces the computational complexity while providing practically the same quality for the restored images as the best existing structures. In the second part of the thesis, the multiscale and multiresolution techniques are put into action to solve some very practical problems, namely the problems of signal denoising and detection of important signal features. The problem of ECG signal denoising is studied in the light of empirical transform-domain Wiener filtering, where ‘empirical’ means that all algorithm parameters are carefully tuned to achieve the best possible performance. First, justification is pro-vided for the way to choose the best decorrelating (essentially multiscale) transform. Then, suggestions regarding interpreting and manipulating dependences of the intra-scale and inter-scale coefficients are given based on deep theoretical understanding and extensive experimental work. The elaborated final algorithm seems to be one of the best known in the field. The problem of closed glottis interval determination is solved by means of a multiresolution analysis. The decomposition involves linear phase octave band filters with sufficient time-frequency localization, most of them representing wavelet bases, but also some half-band filters have been considered. In fact, the half-band filters (not constrained by perfect reconstruction, hence shorter) showed good performance. This put the focus on the multiresolution structure which seems to be the most promising tool for this task. http://sp.cs.tut.fi/cgi-bin/cgiwrap/spwww/thesis.cgi?id=92 Merta, Marja - M.Sc. thesis, 17.09.2003, Avainsanat: geneettinen algoritmi, sekventiaalinen data, business intelligence. Yritys saa kilpailijoihinsa liittyvää dataa tiedostusvälineistä sekä henkilökunnaltaan. Datana saatuja tapahtumia pyritään analysoimaan, ja saamaan näin tietoa sen hetkisestä tilanteesta. Datan luonne vaikeuttaa kuitenkin analysointia, sillä data harvoin on tarkkaa, täydellistä ja varmaa. Tässä diplomityössä on kehitetty menetelmä uutisvirrasta saatavien tapahtumien muodostaman tapahtumasekvenssin tulkitsemiseen. Tulkinta perustuu käyttäjän tallentamiin malleihin. Mallit ovat tapahtumien muodostamia sekvenssejä, joilla on jokin korkeamman tason merkitys. Menetelmän pohjana on käytetty geneettistä algoritmia, jonka avulla tapahtumasekvenssi pyritään optimoimaan mahdollisimman hyvin malleja mukaileviin osasekvensseihin. Tapahtumasekvenssin tapahtumat luokitellaan entuudestaan tunnettuihin luokkiin, jotka kuvaavat tapahtumien tyyppejä. Tapahtumien tyyppien lisäksi oleellista on niiden tapahtumajärjestys. Tämä kaksitasoisuus vaikeuttaa geneettisen algoritmin soveltamista, koska sekä tapahtuma- että aikataso on otettava huomioon algoritmin toiminnassa. Kaksitasoisuuden vuoksi on ollut välttämätöntä muokata tapaa, jolla geneettisen algoritmin kustannuksen laskenta, risteytys ja mutatointi yleensä tehdään. Menetelmän pohjalta on ohjelmoitu sovellus, jota on testattu. Ohjelma sietää varsin hyvin sisäänmenodatassa olevia virheitä, ja saadut tulokset ovat rohkaisevia. Niinpä menetelmää ja ohjelmaa aiotaan jatkokehittää. http://sp.cs.tut.fi/cgi-bin/cgiwrap/spwww/thesis.cgi?id=93 Monagas Machin, Vidina - M.Sc. thesis, 17.09.2003, (no abstract) http://sp.cs.tut.fi/cgi-bin/cgiwrap/spwww/thesis.cgi?id=94 Salko, Eija - M.Sc. thesis, 17.09.2003, (no abstract) http://sp.cs.tut.fi/cgi-bin/cgiwrap/spwww/thesis.cgi?id=95 Toivonen, Jarmo - Licensiate thesis, 17.09.2003, (no abstract) http://sp.cs.tut.fi/cgi-bin/cgiwrap/spwww/thesis.cgi?id=96 Havukainen, Kai - M.Sc. thesis, 15.10.2003, (no abstract) http://sp.cs.tut.fi/cgi-bin/cgiwrap/spwww/thesis.cgi?id=97 Järvinen, Roope - M.Sc. thesis, 15.10.2003, (no abstract) http://sp.cs.tut.fi/cgi-bin/cgiwrap/spwww/thesis.cgi?id=98 Koppinen, Konsta - Doctoral thesis, 15.10.2003, (no abstract) http://sp.cs.tut.fi/cgi-bin/cgiwrap/spwww/thesis.cgi?id=99 Nivala, Juhana - M.Sc. thesis, 15.10.2003, (no abstract) http://sp.cs.tut.fi/cgi-bin/cgiwrap/spwww/thesis.cgi?id=100 Halinen, Sami - M.Sc. thesis, 12.11.2003, (no abstract) http://sp.cs.tut.fi/cgi-bin/cgiwrap/spwww/thesis.cgi?id=101 Koivistoinen, Heidi - M.Sc. thesis, 12.11.2003, (no abstract) http://sp.cs.tut.fi/cgi-bin/cgiwrap/spwww/thesis.cgi?id=102 Koskinen, Toni - M.Sc. thesis, 12.11.2003, (no abstract) http://sp.cs.tut.fi/cgi-bin/cgiwrap/spwww/thesis.cgi?id=103 Salonen, Markku - M.Sc. thesis, 12.11.2003, (no abstract) http://sp.cs.tut.fi/cgi-bin/cgiwrap/spwww/thesis.cgi?id=104 Hautaniemi, Sampsa - Doctoral thesis, 10.12.2003, (no abstract) http://sp.cs.tut.fi/cgi-bin/cgiwrap/spwww/thesis.cgi?id=105 Kunttu, Iivari - Licensiate thesis, 10.12.2003, (no abstract) http://sp.cs.tut.fi/cgi-bin/cgiwrap/spwww/thesis.cgi?id=106 Landau, David - M.Sc. thesis, 10.12.2003, (no abstract) http://sp.cs.tut.fi/cgi-bin/cgiwrap/spwww/thesis.cgi?id=107 Heittola, Toni - M.Sc. thesis, 14.01.2004, (no abstract) http://sp.cs.tut.fi/cgi-bin/cgiwrap/spwww/thesis.cgi?id=108 Lepistö, Leena - Licensiate thesis, 14.01.2004, The purpose of this licentiate thesis is to find methods for the classification of rock images. The varying properties of rock make rock image classification a demanding task. Also the existing methods for content-based image retrieval were tested in the classification of rock images. However, these methods were not accurate in the classification. Therefore, there was a need to develop practical rock image classification methods, which take into account the special features of rock. The goal of this thesis was to apply the pattern recognition methods for the classification of the rock images. The classification was based on the color and texture properties of rock. The previous work provided a good basis for finding the classification methods. In this work, the selected classifying features were such that they characterize the content of the rock images. These features also proved to be suitable for the classification of non-homogenous images. The classification ability of these features was tested using three image databases. In this work, also the surface cracking of rock was inspected. This was made using surface reflection images of rock plates. The selected features proved to be useful in the classification experiments. Among the texture features, especially Gabor-features and directional histogram gave good classification result using non-homogenous rock textures. In the case of color, histograms and correlograms were successful in the classification of the rock images. Also the surface reflection method in the analysis of the surface cracking proved to be effective. The obtained results show that image analysis can be used in many kinds of applications in the rock and stone inspection. The application fields are for example quality control in rock industry, the construction of regular rock surfaces, and rock research. http://sp.cs.tut.fi/cgi-bin/cgiwrap/spwww/thesis.cgi?id=109 Lozano Sanchez, Tatiana - M.Sc. thesis, 14.01.2004, (no abstract) http://sp.cs.tut.fi/cgi-bin/cgiwrap/spwww/thesis.cgi?id=110 Lahti, Johannes - M.Sc. thesis, 10.03.2004, (no abstract) http://sp.cs.tut.fi/cgi-bin/cgiwrap/spwww/thesis.cgi?id=111 Penttilä, Juuso - M.Sc. thesis, 10.03.2004, (no abstract) http://sp.cs.tut.fi/cgi-bin/cgiwrap/spwww/thesis.cgi?id=112 Ryynänen, Matti - M.Sc. thesis, 10.03.2004, (no abstract) http://sp.cs.tut.fi/cgi-bin/cgiwrap/spwww/thesis.cgi?id=113 Klapuri, Anssi - Doctoral thesis, 17.03.2004, Signal processing methods for the automatic transcription of music are developed in this thesis. Music transcription is here understood as the process of analyzing a music signal so as to write down the parameters of the sounds that occur in it. The applied notation can be the traditional musical notation or any symbolic representation which gives sufficient information for performing the piece using the available musical instruments. Recovering the musical notation automatically for a given acoustic signal allows musicians to reproduce and modify the original performance. Another principal application is structured audio coding: a MIDI-like representation is extremely compact yet retains the identifiability and characteristics of a piece of music to an important degree. The scope of this thesis is in the automatic transcription of the harmonic and melodic parts of real-world music signals. Detecting or labeling the sounds of percussive instruments (drums) is not attempted, although the presence of these is allowed in the target signals. Algorithms are proposed that address two distinct subproblems of music transcription. The main part of the thesis is dedicated to multiple fundamental frequency (F0) estimation, that is, estimation of the F0s of several concurrent musical sounds. The other subproblem addressed is musical meter estimation. This has to do with rhythmic aspects of music and refers to the estimation of the regular pattern of strong and weak beats in a piece of music. For multiple-F0 estimation, two different algorithms are proposed. Both methods are based on an iterative approach, where the F0 of the most prominent sound is estimated, the sound is cancelled from the mixture, and the process is repeated for the residual. The first method is derived in a pragmatic manner and is based on the acoustic properties of musical sound mixtures. For the estimation stage, an algorithm is proposed which utilizes the frequency relationships of simultaneous spectral components, without assuming ideal harmonicity. For the cancelling stage, a new processing principle, spectral smoothness, is proposed as an efficient new mechanism for separating the detected sounds from the mixture signal. The other method is derived from known properties of the human auditory system. More specifically, it is assumed that the peripheral parts of hearing can be modelled by a bank of bandpass filters, followed by half-wave rectification and compression of the subband signals. It is shown that this basic structure allows the combined use of time-domain periodicity and frequency-domain periodicity for F0 extraction. In the derived algorithm, the higher-order (unresolved) harmonic partials of a sound are processed collectively, without the need to detect or estimate individual partials. This has the consequence that the method works reasonably accurately for short analysis frames. Computational efficiency of the method is based on calculating a frequency-domain approximation of the summary autocorrelation function, a physiologically-motivated representation of sound. Both of the proposed multiple-F0 estimation methods operate within a single time frame and arrive at approximately the same error rates. However, the auditorily-motivated method is superior in short analysis frames. On the other hand, the pragmatically-oriented method is "complete" in the sense that it includes mechanisms for suppressing additive noise (drums) and for estimating the number of concurrent sounds in the analyzed signal. In musical interval and chord identification tasks, both algorithms outperformed the average of ten trained musicians. For musical meter estimation, a method is proposed which performs meter analysis jointly at three different time scales: at the temporally atomic tatum pulse level, at the tactus pulse level which corresponds to the tempo of a piece, and at the musical measure level. Acoustic signals from arbitrary musical genres are considered. For the initial time-frequency analysis, a new technique is proposed which measures the degree of musical accent as a function of time at four different frequency ranges. This is followed by a bank of comb filter resonators which perform feature extraction for estimating the periods and phases of the three pulses. The features are processed by a probabilistic model which represents primitive musical knowledge and performs joint estimation of the tatum, tactus, and measure pulses. The model takes into account the temporal dependencies between successive estimates and enables both causal and noncausal estimation. In simulations, the method worked robustly for different types of music and improved over two state-of-the-art reference methods. Also, the problem of detecting the beginnings of discrete sound events in acoustic signals, onset detection,is separately discussed. http://sp.cs.tut.fi/cgi-bin/cgiwrap/spwww/thesis.cgi?id=114 Alaya Cheikh, Faouzi - Doctoral thesis, 06.04.2004, The thesis considers different aspects of the development of a system called MUVIS developed in the MuVi project for content-based indexing and retrieval in large image databases. The issues discussed are system design, graphical user interface design, shape features extraction and their corresponding similarity measures, and the use of relevance feedback to enhance the retrieval performance. Query by content, or content-based retrieval has recently been proposed as an alternative to text-based retrieval for media such as images, video and audio. Text-based retrieval is no longer appropriate for indexing such media, for several reasons. Firstly, keyword annotation is labor intensive, and it is not even possible when large sets of images are to be indexed. Secondly, these annotations are drawn from a predefined set of keywords which cannot cover all possible concepts images may represent. Finally, keywords assignment is subjective to the person making it. Therefore, content-based image retrieval (CBIR) systems propose to index the media documents based on features extracted from their content rather than by textual annotations. For still images, these features can be color, shape, texture, objects layout, edge direction, etc. Query by content has been a very active research field, with many systems proposed by industrial and academic teams. Building such systems requires expertise in different fields of information technology: databases and indexing structures, system design and integration, graphical user interfaces (GUI), signal processing and analysis, man-machine interaction, user psychology, etc. The development of the MUVIS system for content-based image retrieval was a challenging work that required solutions to different problems. Therefore, this thesis contributed to different phases of the system development; including system design, graphical user interface development, feature extraction, similarity measures, and relevance feedback. The system is designed in a client-server architecture. The graphic user interface proposes a new way of visualizing the color features used and permits intuitive interactions with it. Moreover, it allows several ways of query formulation: by example image, by histogram, or by single object or composed image. The second contribution of this thesis is the study of shape features and related similarity measures. These can be classified into two categories: wavelet-based approaches and ordinal correlation based ones. The former is a set of contourbased descriptors, where the features are extracted at significant topological points of the objects contours, which are the high curvature points (HCP); detected based on the local maxima of the wavelet transform of the 1D contour representation. The retrieval results are compared to those of the curvature scale space (CSS) and to a ranked list of images by a group of human users. The results show that our proposed schemes have comparable or better results than the CSS. Moreover, the proposed algorithms are computationally more efficient than the CSS, and are not sensitive to noise corruption or partial occlusion of the contour. The latter approach is a region-based technique, which applies the ordinal correlation framework on the 2D distance transform of binary contour image to extract the shape features. The distance transformation disseminates the information highly concentrated in the binary contour into the surrounding pixels, making the approach less sensitive to small variation of the contour caused by noise, imperfect segmentation or small object alignment errors. Being region-based, this approach can be applied to simple contours as well as shapes composed of several independent regions which may even contain internal holes. The third contribution of this thesis consists of the implementation of a relevance feedback (RF) scheme in our CBIR system. RF techniques allow the CBIR systems to learn from the user interaction and enhance their performance in subsequent retrieval iterations. The proposed technique combines both re-weighting of the feature vector entries and adaptive resolution of the feature vectors to achieve a higher performance in discriminating relevant and irrelevant objects for a given query. http://sp.cs.tut.fi/cgi-bin/cgiwrap/spwww/thesis.cgi?id=115 Ahmaniemi, Ali - M.Sc. thesis, 07.04.2004, (no abstract) http://sp.cs.tut.fi/cgi-bin/cgiwrap/spwww/thesis.cgi?id=116 Mäki-Kullas, Mikko - M.Sc. thesis, 07.04.2004, (no abstract) http://sp.cs.tut.fi/cgi-bin/cgiwrap/spwww/thesis.cgi?id=117 Lohan, Florin - Doctoral thesis, 03.06.2004, (no abstract) http://sp.cs.tut.fi/cgi-bin/cgiwrap/spwww/thesis.cgi?id=118 Tunyan, Knarik - Doctoral thesis, 24.09.2004, One of the most important objectives in solving practical problems is design of computationally efficient and stable algorithms. A significant part of signal processing employs a variety of numerical methods of linear algebra and linear programming, for instance, inverting matrices, solving for systems of linear equations, least squares problems, and optimization problems. In this thesis, based on the concept of partial orthogonality, some fundamental methods in linear algebra and linear programming are revised. Namely, the QR decomposition is generalized; new methods in linear algebra are developed, which can provide computational savings and/or allow obtaining an improved solution, compared with the existing algorithms; new pivot rules for the simplex method of interior points are proposed. Considering a few important applications in signal processing, the effectiveness of this theoretical results in practice is illustrated. Specifically, for the problem of finding both accurate and sparse solution improved results can be obtained, compared to the existing methods. Our approach to the linear least squares problem allows decomposing it into simpler subproblems, yielding computational efficiency in comparison with the algorithms using the Gram-Schmidt and the Householder transformations. As for linear programming, the simplex method of interior points is investigated, which combines the desirable theoretical properties of interior point methods and practical advantages of the simplex method. New pivot rules are proposed. Effectiveness of these rules is also illustrated. The proposed efficient techniques can certainly be used in a wide range of important practical applications, that are addressed within the thesis scope as well. http://sp.cs.tut.fi/cgi-bin/cgiwrap/spwww/thesis.cgi?id=119 Dabov, Kostadin - M.Sc. thesis, 16.05.2006, In this thesis, we develop two novel image-denoising methods based on locally adaptive transforms. The basic idea is to denoise local patches of the image by shrinking (e.g. hard-thresholding, Wiener filtering) the coefficients of a local transform applied on these patches. The local transform aims at decorrelating the true signal so that its energy is compacted in few high-magnitude coefficients that are preserved or only slightly modified by the shrinkage. The rest of the coefficients, which are mainly due to noise, are substantially attenuated or discarded. An inverse transform reconstructs local estimates of the signal within the local patches. Because local estimates that correspond to neighboring points are, in general, overlapping, they form an overcomplete representation of the signal. Hence, we compute the final estimate of an image as a weighted average of the local ones where they overlap. The first method is based on local shape-adaptive transforms. We use the anisotropic LPA-ICI technique to define adaptively the shapes of the local patches for each processed location. Hence, the patches are adaptive with respect to the structures of the true image, such as edges and singularities. This ensures a high level of homogeneity in the signal within the patches, so that a shape-adaptive transform achieves good energy compaction. The second method is based on denoising in 3D transform domain by combining sliding-window processing with block-matching. We utilize the block-matching concept by searching for blocks that are similar to the currently processed one. The matched blocks are stacked together to form a 3D array. The correlation between spatially neighboring pixels and the similarity among matched blocks result in improved energy compaction of a 3D transform that we apply on this array. A subsequent inverse 3D transform yields local estimates of all matched blocks. By improving the energy compaction of the local transform-domain representation of image patches, both methods increase the effectiveness of the shrinkage, when compared to approaches based on non-adaptive transforms. The experimental results show that our techniques deliver state-of-the-art denoising performance in terms of both peak signal-to-noise ratio and subjective visual quality. http://sp.cs.tut.fi/cgi-bin/cgiwrap/spwww/thesis.cgi?id=120 Yang, Ruixing - M.Sc. thesis, 01.03.2006, Correspondence searching and efficiency are two of most important problems that hinder the development of stereo vision and further research of 3D object construction. The current approaches for stereo map generation can be classified as feature-based, area-based and phase-based approaches. Each of them has its own advantages and disadvantages. For feature-based approach, the correspondence searching is relatively precise, however usually the number of detected valid features is not big enough to construct an image depth map. On the other hand, for area-based approach which is usually based on the correlation minimization, the high precision cannot be obtained easily. And for phase-based approach, if there is big change of scale, rotation or movement, it is very difficult to obtain the final correct disparity result. In this thesis, a hybrid algorithm for stereo map generation is investigated. It combines the properties of the feature-based and area-based approaches to achieve better performance. In the algorithm, feature correspondence searching is utilized in a precise manner. First, a corner detector is used to detect the edges and corners in the sensed image and reference image. From the detected features, points that generate wrong correspondences are dropped. The remaining ones are quite robust to be applied to area-based approach for searching correspondence. Then we use Scale Invariant Feature Transform (SIFT) detector to derive a set of SIFT keys. Since SIFT features are multi-dimensional ones, correspondence matching using them would be more robust than matching using feature points such as edges and corners. With the correspondence of the detected features, it is required to interpolate the disparity map obtained from the matched features to generate the complete disparity map. In the algorithm, segmentation based on histogram is proposed for improving the interpolation result. The proposed algorithm has the properties of both classical feature-based and area-based approaches. The selected SIFT feature correspondence searching is very robust and efficient. For edges and corners, some points are dropped to avoid the error that is easily generated by the area-based approach. For each step of the proposed algorithm, the robustness and efficiency are carefully considered, leading to an improved final result. http://sp.cs.tut.fi/cgi-bin/cgiwrap/spwww/thesis.cgi?id=121 Essaky Sankaran, Harish - M.Sc. thesis, 01.03.2007, Image super-resolution refers to the reconstruction of a high resolution image from a set of blurred and noisy low resolution images which are sub-pixel shifted from each other. Each low resolution image contains new information about the scene and super-resolution aims at combining these to give a higher resolution image. This thesis deals with signal reconstruction on higher resolution uniform grid from samples on non-uniform grid. Reconstructing a signal from its non-uniformly sampled values is an often encountered problem in signal processing and is at the core of super-resolution reconstruction, since the sub-pixel shifted images, put together, form a non-uniform grid with respect to the desired super-resolution grid. In this thesis, the problem of reconstructing a signal from its non-uniform samples in shift-invariant spaces is studied and two reconstruction techniques are proposed: a near least squares separable reconstruction technique and a near least squares two-dimensional reconstruction technique, both not previously applied to the super-resolution problem. The separable reconstruction technique is suitable for the special case of interlaced sampling which corresponds to translational motion between the low resolution images. The second technique is an extension of the first one to two dimensions and handles arbitrary motion between the low resolution images. Efficient realizations of both techniques are achieved by digital filtering using the transposed Farrow structure. Both techniques are very computationally efficient and demonstrate good performance compared with other recent resampling methods.