In this paper, we introduce a novel estimator for the instantaneous frequencies of the modes making up multicomponent signals, particularly when they are close or crossing in the time-frequency plane. Our approach is based on locally optimizing the spectrogram window in the time-frequency domain to enhance signal representation, depending on whether the modes are isolated or interfering. To achieve this, we employ an adaptive criterion: for isolated modes, the window is selected to minimize time dependency, while for close or interfering modes, it is chosen to reduce interference. By integrating this adaptive time-frequency representation with ridge detection, and then by using spline fitting, we obtain highly robust instantaneous frequency estimates. Our proposed method demonstrates strong performance in handling MCSs with close or crossing modes, and is shown to outperform state-of-the-art instantaneous frequency estimation techniques, including methods based on synchrosqueezing and the chirplet transform.