This talk describes some recent results of the Signal Processing Group at the Signal Processing laboratory of Tampere University of Technology.
The signal processing research group has five teams: Compression Team, Spectral and Algebraic Methods, Image Processing team, Linear DSP team and Biomedical SP team. In this talk an overview of the teams and their research activities is given. Certain research projects of Compression, Spectral and Image processing teams are discussed in more detail.
One of central problems in filterbank theory is designing optimal compaction gain FIR filters. Starting from the odd polyphase component of the input correlation sequence, which is the only apriori information in the optimal design of a compaction FIR filter of order, we find a suitable extension to an infinite sequence such that the corresponding spectrum is a line spectrum . In quite general conditions it is possible to design the optimal filter such that it has zeros at the line frequencies of and obeys the Nyquist condition, which clearly gives the exact solution for the optimization problem.
In lossles audio compression we consider the use of adaptive-context-based prediction in a sequential mode. We show that lossless compression algorithms with sequential context based prediction can achieve better compression results than with forward-frame-based linear prediction. Two distinct algorithms are proposed and evaluated for audio signals sampled at 48 kHz with 16 bits/sample.
We consider the use of integer wavelet as a decorrelation stage fo adaptive context based lossless audio coding. The original wide band audio signal is first decomposed in wavelet subbands. The resulting coefficients are integer valued and therefore can be transmitted using an adaptivem context based method, in a lossless manner, the decoder being able to reconstruct them and afterwords to perfectly restore the audio waveform.Several ways to encode theinteger wavelet coefficients are explored and the results are compared with those obtained in fullband contex adaptive coding.
In nonlinear filtering we consider a multiresolution pyramidal transform that is based on median operation. This structure can be used in denoising nonstatinary signals. It is shown that for this nonlinear structure it is possible to analytically determine statistical properties. A new method is also considered for window size selection for median filters. The method is based on so called intersecting confidence intervals.
The above and and other algorithms are used in developing methods for context based indexing and retrieval over large image databases. The database work includes an image processing and analysis toolbox, a graphic user interface (in Java and html) and software integration leading ot a full system (called MUVIS).
Jaakko Astola was born in Helsinki, Finland on May 6, 1949. He received the M.Sc. and Ph.D. degrees in mathematics (specialising in error-correcting codes) from Turku University, Finland, in 1973, and 1978 respectively. From 1976 to 1977 he was a research assistant at the Research Institute for Mathematical Sciences of Kyoto University, Kyoto, Japan. Between 1979 and 1987 he was with the Department of Information Technology, Lappeenranta University of Technology, Lappeenranta, Finland, holding various teaching positions in mathematics, applied mathematics and computer science. In 1984 he worked as a visiting scientist in Eindhoven University of Technology, The Netherlands. From 1987 to 1992 he was Associate Professor in Applied Mathematics at Tampere University, Tampere, Finland. Currently he is Professor of Signal Processing at Tampere University of Technology and director of tampere International Center for Signal Processing. He is leading a signal processing research group of about 50 scientists and this group elected in 1999 as a Center of Excellence in Research in Finland by Academy of Finland.
Dr. Astola is being hosted by the Video and Image Processing Laboratory (VIPER) of the School of Electrical and Computer Engineering. Please contact Prof. Edward Delp (ace@ecn.purdue.edu), 765-494-1740 for more information.
Refreshments will be served before the seminar starting at 3:45pm.