ABSTRACT
Fourier Transform (FT) offers a spectral analysis by specifically indicating the frequencies that have signals in addition to their corresponding amplitudes and phases. However, FT generates a time-averaged spectrum, i.e. information from this transform is limited to the instants in which the corresponding frequencies appear. Therefore, FT is not suitable for analysis of non-stationary gestures, i.e. signals whose statistical characteristics evolve as a function of time. Indeed, we cannot have precisions on frequency changes over time since the modulus of the FT of a signal simply produces a temporal average of the spectral content. To overcome this problem, time-frequency representations are used, which is discussed in this chapter. The main methods of time-frequency representations while presenting the continuous and discrete version for each technique as well as its computer implementation are studied.
