Time-frequency analysis (TFA) is an efficient tool to jointly detect point scatterers in SAR images and determine their backscattering properties. In this paper, we focus on TFA in multi-image context (for instance, interferometric stacks or multi-temporal series). A new TFA algorithm is proposed based on ??spectrograms?? which are 4D hyper images representing the behaviour of each pixel with respect to range and azimuth frequencies. Spectrograms are analyzed in a multi-image context: they are combined to select pixels with stable behaviour over time (persistent scatterer detection) or unstable behaviour (change detection). Spectrograms characterize short term variations, whereas, in this paper, we characterize long term variations based on spectrogram properties. Preliminary results obtained with Spotlight interferometric TerraSAR-X images are discussed.