Fluorescence lifetime imaging microscopy (FLIM) can be used to quantify molecular reactions in cells by detecting fluorescence resonance energy transfer (FRET). Confocal FLIM systems based on time correlated single photon counting (TCSPC) methods provide high spatial resolution and high sensitivity, but suffer from poor signal to noise ratios (SNR) that complicate quantitative analysis. We extend a global analysis method, originally developed for single frequency domain FLIM data, with a new filtering method optimized for FRET-FLIM data and apply it to TCSPC data. With this approach, the fluorescent lifetimes and relative concentrations of free and interacting molecules can be reliably estimated, even if the SNR is low. The required calibration values of the impulse response function are directly estimated from the data, eliminating the need for reference samples. The proposed method is efficient and robust, and can be routinely applied to analyze FRET-FLIM data acquired in intact cells.