In the last few years, a gain in adaptive methods to represent a given signal or image arose in the literature. One of the most used but also less understood is the Empirical Mode Decomposition (EMD). After recalling the main concept and properties of the EMD, I will introduce a new way to build adaptive wavelets aiming to perform the same kind of analysis as the EMD. Those wavelets, called empirical wavelets, are based on the idea of detecting modes of compact support in the Fourier domain. I will provide the formulation to build such wavelets and discuss the most delicate part which is the detection of modes in the Fourier domain. Next, I will show that it possible to extend this concept to existing 2D transforms (tensor approach, Littlewood-Paley, Ridgelets and Curvelets). Finally, I will present preliminary results in the analysis of electroencephalogram signals and give ideas of future investigation both on the theoretical and application sides.