Fuzzy Multiresolution Signal Processing (FMSP)

摘要

In many different types of signals, like radar signals, biomedical signals (e.g. ECG, EEG, Echocardiographic images, etc.), and Chemometric signals (e.g. Infra-Red spectra, Raman spectra, Chromatographic spectra, etc.) most of the interesting information is carried by singularities caused by sharp signal changes such as peaks and edges. Detection and locating these singularities is, therefore, a general goal in many signal and image processing problems. This paper describes a new Fuzzy Multiresolution Signal Processing (FMSP) technique for singularity locating in a signal. The Wavelet Transform is employed to produce the multiresolution representation of the input signal. The information at various signal resolutions is then fuzzified using appropriate fuzzy membership functions, where each resulting fuzzy subset describes to what degree any pixel in the signal domain can coincident with the position of the singularity. The final result is obtained by combining the information in the various fuzzy subsets of the signal. The paper demonstrates application of the FMSP technique to the synthetic and real signals and compares its performance to those of the conventional gradient and multiresolution based techniques.