Time-frequency decomposition for analysis and retrieval of 1-D signals

The time-frequency decomposition of wavelets is used to characterize 1D signals: the goal is to associate to each discrete signal a set of descriptors for fast retrieval in large database and/or recognition in noisy environment. Experiments are shown for an heterogeneous set of one dimensional signals, both synthetic and natural (audio signals). The efficacy of the method is tested against corruption by (white and colored) noise and/or attenuation, in order to simulate the transmission on noisy channels and LANs. Also the extension of the algorithm to 2D signals shows satisfactory results and it has been applied in a content based retrieval engine for image databases