Projects / Programmes
Decomposition Models for Mono- and Two-Dimensional Compound Signals
Code |
Science |
Field |
Subfield |
2.06.06 |
Engineering sciences and technologies |
Systems and cybernetics |
Related areas |
Code |
Science |
Field |
T111 |
Technological sciences |
Imaging, image processing |
T121 |
Technological sciences |
Signal processing |
Signal Processing, Image Processing, MISO and MIMO Modelling, Time-Frequency Representations, Blind Deconvolution, Cellular Automata, Cellular Neural Networks, Biomedical Signal and Image Analysis.
Researchers (2)
no. |
Code |
Name and surname |
Research area |
Role |
Period |
No. of publicationsNo. of publications |
1. |
12156 |
PhD Danilo Korže |
Systems and cybernetics |
Researcher |
1998 - 1999 |
209 |
2. |
08061 |
PhD Damjan Zazula |
Systems and cybernetics |
Head |
1998 - 1999 |
789 |
Organisations (1)
Abstract
The main project objective is the development of decomposition models of the MISO and MIMO types, aimed at the analysis of compound signals. The common decomposition basis is searched for through the approaches of cepstral analysis, higher-order statistics, parametric search techniques, time-scale representation, Hough transform, cellular neural networks, and some others. Theoretic derivation is strongly accompanied by applications to the biomedical signals and images, like ECG, EMG, and ultrasonic recordings.
The research is conducted over three years in two parallel streams. The first one is dealing with mono-dimensional signals, whereas the second one tackles the two-dimensional signals, i.e. images. Both groups have the following common activities: unified conceptual MISO and MIMO models, development of the decomposition methods regarding these models, realisation of the computer algorithms, construction of computer simulators in order to verify the derived decomposition approaches, application of the new algorithms to the real signal recordings and images.
So far, three approaches have been elaborated for mono-dimensional signals: multimodal parametric search realised through iterative monomodal descents (verified on synthetic and real needle and surface EMGs), bicepstral decomposition built on a novel approach with asymptotically exact computation of differential cepstrum (verified on synthetic surface EMGs), and time-scale phase analysis and representation for an improved accuracy of detection of the events in time (verified on synthetic and real surface EMGs). In the two-dimensional domain, Hough transform has been tested and eliminated as inappropriate from the standpoint of the project goals. However, two other approaches with cellular automata and cellular neural networks proved beneficial and capable of decomposing rather difficult ultrasonic images (clinical up-takes of women’s ovaries). Further improvements are expected by inclusion of the decomposition models into the neural network learning procedures.