Clustering text documents is a fundamental task in modern data analysis, requiring approaches which perform well both in terms of solution quality and computational efficiency. Spherical k-means clustering is one approach to address both issues, employing cosine dissimilarities to perform prototype-based partitioning of term weight representations of the documents. This paper presents the theory underlying the standard spherical k-means problem and suitable extensions, and introduces the R extension package skmeans which provides a computational environment for spherical k-means clustering featuring several solvers: a fixed-point and genetic algorithm, and interfaces to two external solvers (CLUTO and Gmeans). Performance of these solvers is investigated by means of a large scale benchmark experiment. (authors' abstract)

Multi-resolution analysis has been a very popular technique in the recent years. Wavelets have been used extensively to perform multi resolution image expansion and analysis. DCT, however, has been used to compress image but not for multi resolution image analysis. This thesis is an attempt to explore the possibilities of using DCT for multi-resolution image analysis. Naive implementation of block DCT for multi-resolution expansion has many difficulties that lead to signal distortion. One of the main causes of distortion is the blocking artifacts that appear when reconstructing images transformed by DCT. The new algorithm is based on line DCT which eliminates the need for block processing. The line DCT is one dimensional array based on cascading the image rows and columns in one transform operation. Several images have been used to test the algorithm at various resolution levels. The reconstruction mean square error rate is used as an indication to the success of the method. The proposed algorithm has also been tested against the traditional block DCT.

No description available.

Yes / New orthogonal wavelet transforms can be designed by changing the wavelet basis functions or by constructing new low-pass filters (LPF). One family of wavelet may appeal, in use, to a particular application than another. In this study, the wavelet transform based on raisedcosine spectrum is used as an independent orthogonal wavelet to study multicarrier modulation behaviour over multipath channel environment. Then, the raised-cosine wavelet is compared with other well-known orthogonal wavelets that are used, also, to build multicarrier modulation systems. Traditional orthogonal wavelets do not have side-lobes, while the raised-cosine wavelets have lots of side-lobes; these characteristics influence the wavelet behaviour. It will be shown that the raised-cosine wavelet transform, as an orthogonal wavelet, does not support the design of multicarrier application well like the existing well-known orthogonal wavelets.

Discrete Cosine Transform (DCT) is widely used transform in many areas of the current information age. It is used in signal compression such as voice recognition, shape recognition and also in FBI finger prints. DCT is the standard compression system used in JPEG format. The DCT quality deteriorates at low-bit compression rate. The deterioration is due to the blocking artifact inherent in block DCT. One of the successful attempts to reduce these blocking artifacts was conversion of Block-DCT into Line-DCT. In this thesis we will explore the Line-DCT and introduce a new form of line-DCT called Bidirectional-DCT, which retains the properties of Line- DCT while improving computational efficiency. The results obtained in this thesis show significant reduction in processing time both in one dimensional and two dimensional DCT in comparison with the traditional Block-DCT. The quality analysis also shows that the least mean square error is considerably lower than the traditional Block-DCT which is a consequence of removing the blocking artifacts. Finally, unlike the traditional block DCT, the Bidirectional-DCT enables compression with very low bit rates and very low blocking artifacts.

We propose an efficient iterative data interpolation method using continuity along reflectors in seismic images via curvelet and discrete cosine transforms. The curvelet transform is a new multiscale transform that provides sparse representations for images that comprise smooth objects separated by piece-wise smooth discontinuities (e.g. seismic images). The advantage of using curvelets is that these frames are sparse for high-frequency caustic-free solutions of the wave-equation. Since we are dealing with less than ideal data (e.g. bandwidth-limited), we compliment the curvelet frames with the discrete cosine transform. The latter is motivated by the successful data continuation with the discrete Fourier transform. By choosing generic basis functions we circumvent the necessity to make parametric assumptions (e.g. through linear/parabolic Radon or demigration) regarding the shape of events in seismic data. Synthetic and real data examples demonstrate that our algorithm provides interpolated traces that accurately reproduce the wavelet shape as well as the AVO behavior along events in shot gathers.

The master's thesis focuses on the study of algorithms dealing with noise separation from musical signal. The first chapter is an introduction into methods which are used for noise removal of the musical signal. Furthermore, this chapter describes theory to the issue, specifically a description of transformations for converting from time to frequency domain, and finally thresholding method of spectral coefficients is explained in detail. The aim of the second chapter is an analysis of the proposed algorithm, which is engaged in testing. From the beginning fast algorithms of gradual transformation are described and then a detailed description of the algorithm as a whole. Later, this chapter deals with the selection of audio recordings and with preparation of these recordings for the actual testing. Finally, testing of audio samples is presented in the third chapter of this thesis. This chapter also concludes comparison of individual transformations, achieved results and review of algorithm.

The expression ''Hire for attitude, train for skills'' is used as a motive to create a matching program where companies and job seekers' soft qualities are measured and compared against each other. Are there better or worse methods for this purpose, and how do they compare with each other? By associating soft qualities with companies and job seekers, it is possible to generate a value for how well they match. Therefore, data has been collected on several companies and job seekers. Their associated qualities are then translated into numerical vectors that can be used for matching purposes, where vectors closer together are more equal than vectors with greater distances. When it comes to analyzing and comparing the qualities, several methods have been used and compared with a subsequent discussion about their suitability. One consequence of the lack of a proper standard for presenting the qualities of companies and job seekers is that the data is messy and varied. An expected conclusion from the result is that the most flexible method is the one that generates the most accurate results.

Yes / Today, computer based face recognition is a mature and reliable mechanism which is being practically utilised for many access control scenarios. As such, face recognition or authentication is predominantly performed using ‘perfect’ data of full frontal facial images. Though that may be the case, in reality, there are numerous situations where full frontal faces may not be available — the imperfect face images that often come from CCTV cameras do demonstrate the case in point. Hence, the problem of computer based face recognition using partial facial data as probes is still largely an unexplored area of research. Given that humans and computers perform face recognition and authentication inherently differently, it must be interesting as well as intriguing to understand how a computer favours various parts of the face when presented to the challenges of face recognition. In this work, we explore the question that surrounds the idea of face recognition using partial facial data. We explore it by applying novel experiments to test the performance of machine learning using partial faces and other manipulations on face images such as rotation and zooming, which we use as training and recognition cues. In particular, we study the rate of recognition subject to the various parts of the face such as the eyes, mouth, nose and the cheek. We also study the effect of face recognition subject to facial rotation as well as the effect of recognition subject to zooming out of the facial images. Our experiments are based on using the state of the art convolutional neural network based architecture along with the pre-trained VGG-Face model through which we extract features for machine learning. We then use two classifiers namely the cosine similarity and the linear support vector machines to test the recognition rates. We ran our experiments on two publicly available datasets namely, the controlled Brazilian FEI and the uncontrolled LFW dataset. Our results show that individual parts of the face such as the eyes, nose and the cheeks have low recognition rates though the rate of recognition quickly goes up when individual parts of the face in combined form are presented as probes.

Le suivi de partition est défini comme étant la synchronisation sur ordinateur entre une partition musicale connue et le signal sonore de l'interprète de cette partition. Dans le cas particulier de la voix chantée, il y a encore place à l'amélioration des algorithmes existants, surtout pour le suivi de partition en temps réel. L'objectif de ce projet est donc d'arriver à mettre en oeuvre un logiciel suiveur de partition robuste et en temps-réel utilisant le signal numérisé de voix chantée et le texte des chansons. Le logiciel proposé utilise à la fois plusieurs caractéristiques de la voix chantée (énergie, correspondance avec les voyelles et nombre de passages par zéro du signal) et les met en correspondance avec la partition musicale en format MusicXML. Ces caractéristiques, extraites pour chaque trame, sont alignées aux unités phonétiques de la partition. En parallèle avec cet alignement à court terme, le système ajoute un deuxième niveau d'estimation plus fiable sur la position en associant une segmentation du signal en blocs de chant à des sections chantées en continu dans la partition. La performance du système est évaluée en présentant les alignements obtenus en différé sur 3 extraits de chansons interprétés par 2 personnes différentes, un homme et une femme, en anglais et en français.