The Empirical Wavelet Transform is an approach proposed to decompose a signal accordingly to its contained information, addressing the Empirical Mode Decomposition algorithm’s issues of nonlinearity and lack of mathematical theory. However, in the extension of the EWT algorithms into 2D, the wavelets are still based on a prescribed type of partitioning of the Fourier domain. In this thesis, we investigate the opportunity to rid the prescribed partitioning by first implementing scale-space theory inspired techniques to detect the position of the 2D harmonic modes, then partitioning the Fourier domain into Voronoi cells that will serve as supports to build empirical wavelet filters. We show the efficiency of the proposed approach in extracting different harmonic modes.
