Improvement on joint time-frequency representation with application of image processing technique

Time-frequency representation makes it possible to analyse non-stationary signal, which traditional Fourier transform is not adapted to. However, the trade-off between time and frequency accuracy is inevitable because of the constraint of Heisenberg-Gabor inequality. Current methods such as adaptive kernel design can give satisfactory result for some particular signals, but either the prior knowledge or parameter search is needed. This paper is intended to discuss the application of image processing technique to time-frequency distribution, which can effectively improve the resolution of representation and make the detection and estimation of signal an easy job.