In this study, time-frequency representation (TFR) of a multiple-component signal is treated as an image and the segmentation aims at separating and extracting time-frequency signatures suitable for further feature extraction. The main difficulties in this case arise from the complexity of the signal, presence of noise and restricted localization properties of the time-frequency distributions. Our study addresses these issues by combining watershed image segmentation, local feature extraction and a region growing procedure to group subsets of the image into meaningful structures. The performance of our method is illustrated with the examples where very limited a priori knowledge of the signal characteristics is available
|Translated title of the contribution||Region feature-based segmentation of time-frequency images|
|Title of host publication||11th International Workshop on Systems, Signals and Image Processing. Ambient Multimedia, Poznan, Poland|
|Publisher||Polish Society for Theoretical and Applied Electrical Engineering|
|Pages||375 - 378|
|Number of pages||4|
|Publication status||Published - 13 Sep 2004|
Loza, AT., Canagarajah, CN., & Bull, DR. (2004). Region feature-based segmentation of time-frequency images. In 11th International Workshop on Systems, Signals and Image Processing. Ambient Multimedia, Poznan, Poland (pp. 375 - 378). Polish Society for Theoretical and Applied Electrical Engineering.